Lamapi commited on
Commit
82dfc77
·
verified ·
1 Parent(s): 641783a

Publish Next-Match model

Browse files
Files changed (15) hide show
  1. .gitattributes +2 -0
  2. README.md +64 -0
  3. catalog.json +1 -0
  4. catalog.png +3 -0
  5. combined.safetensors +3 -0
  6. evaluation.json +1045 -0
  7. image.safetensors +3 -0
  8. inference.py +55 -0
  9. match_core.py +79 -0
  10. model.json +0 -0
  11. nlrt.py +226 -0
  12. preview.png +3 -0
  13. render.py +66 -0
  14. selftest.json +0 -0
  15. text.safetensors +3 -0
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ catalog.png filter=lfs diff=lfs merge=lfs -text
37
+ preview.png filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: pytorch
4
+ tags:
5
+ - composed-image-retrieval
6
+ - multimodal
7
+ - turkish
8
+ - next-lab
9
+ language:
10
+ - tr
11
+ - en
12
+ ---
13
+
14
+ # Next Match
15
+
16
+ Next Match finds a catalog item from a **reference image plus a short edit in Turkish or English** —
17
+ *"bunun kırmızı olanı"*, *"çizgili olanı"*, *"same but a hat"*. The test catalog is built so that the image alone (it
18
+ finds look-alikes in the wrong colour) and the text alone (it finds red things of any kind) both fail; the fused query
19
+ succeeds. The catalog is **synthetic**: procedurally drawn product icons (8 categories × 8 colours × 3 patterns).
20
+
21
+ **Space:** https://huggingface.co/spaces/thelamapi/Next-Match — pick a product (or upload a picture), type a change, and compare the three rankings live.
22
+
23
+ ## Results (1600 test queries, 192-item catalog, unseen renderings)
24
+
25
+ Recall@1 / Recall@5:
26
+
27
+ | Query group | n | Image + text (Next Match) | Image only | Text only |
28
+ |---|---:|---:|---:|---:|
29
+ | all | 1600 | 0.932 / 0.966 | 0.000 / 0.316 | 0.054 / 0.263 |
30
+ | edit=color | 648 | 0.994 / 1.000 | 0.000 / 0.358 | 0.051 / 0.218 |
31
+ | edit=pattern | 311 | 0.836 / 0.897 | 0.000 / 0.878 | 0.016 / 0.074 |
32
+ | edit=category | 322 | 1.000 / 1.000 | 0.000 / 0.000 | 0.031 / 0.183 |
33
+ | edit=color+pattern | 319 | 0.834 / 0.931 | 0.000 / 0.000 | 0.122 / 0.621 |
34
+ | lang=tr | 747 | 0.995 / 1.000 | 0.000 / 0.317 | 0.051 / 0.258 |
35
+ | lang=en | 853 | 0.878 / 0.937 | 0.000 / 0.314 | 0.057 / 0.267 |
36
+ | template=seen | 1104 | 1.000 / 1.000 | 0.000 / 0.323 | 0.057 / 0.251 |
37
+ | template=held-out | 496 | 0.782 / 0.891 | 0.000 / 0.298 | 0.048 / 0.290 |
38
+ | combo=seen | 1192 | 0.933 / 0.966 | 0.000 / 0.348 | 0.057 / 0.296 |
39
+ | combo=held-out | 408 | 0.931 / 0.966 | 0.000 / 0.221 | 0.047 / 0.167 |
40
+
41
+ *template=held-out*: phrasings never used in training (e.g. "show me the red one", "kırmızı olanını göster").
42
+ *combo=held-out*: the reference or target is one of 24 attribute combinations never rendered during training.
43
+
44
+ ![Test queries and Recall@K](preview.png)
45
+
46
+ ## Method
47
+ * Image encoder: 4 conv blocks (16/32/64/64 channels) + linear → 64-d. Catalog keys are L2-normalised image features.
48
+ * Text: Turkish-aware lower-casing and longest-prefix stemming over a 61-word vocabulary (so "kırmızısı" or "olanını" still match), mean word embedding → MLP.
49
+ * Fusion: q = normalise(f_img + MLP([f_img, f_text])) — the text predicts a residual edit of the image feature.
50
+ * Baselines: image-only (image encoder + MLP head) and text-only (text encoder vs its own image keys), same architecture budget, 6000 steps, batch 128, InfoNCE with τ = 0.1.
51
+ * The Space runs all three encoders with `nlrt.js` on exported weights; Node parity max abs difference 1.6653345369377348e-16.
52
+
53
+ ## Usage
54
+ ```python
55
+ from inference import load, search
56
+ spec = load()
57
+ ids = search(spec, ref_image, "bunun kırmızı olanı") # ref_image: (48, 48, 3) uint8; ids index catalog.json
58
+ ```
59
+
60
+ ## Limitations
61
+ * Toy domain: flat icons on plain backgrounds. Real product photos need a pretrained vision encoder and real catalog data.
62
+ * Vocabulary is closed: words outside it are ignored (the Space shows which words were understood).
63
+ * Uploaded photos are far outside the training distribution and are a curiosity, not a supported use.
64
+ * Trained on NVIDIA L4; profile full. Failure examples are in `evaluation.json`.
catalog.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"cols": 16, "items": [{"category": "tshirt", "color": "red", "pattern": "plain"}, {"category": "tshirt", "color": "red", "pattern": "striped"}, {"category": "tshirt", "color": "red", "pattern": "dotted"}, {"category": "tshirt", "color": "blue", "pattern": "plain"}, {"category": "tshirt", "color": "blue", "pattern": "striped"}, {"category": "tshirt", "color": "blue", "pattern": "dotted"}, {"category": "tshirt", "color": "green", "pattern": "plain"}, {"category": "tshirt", "color": "green", "pattern": "striped"}, {"category": "tshirt", "color": "green", "pattern": "dotted"}, {"category": "tshirt", "color": "yellow", "pattern": "plain"}, {"category": "tshirt", "color": "yellow", "pattern": "striped"}, {"category": "tshirt", "color": "yellow", "pattern": "dotted"}, {"category": "tshirt", "color": "black", "pattern": "plain"}, {"category": "tshirt", "color": "black", "pattern": "striped"}, {"category": "tshirt", "color": "black", "pattern": "dotted"}, {"category": "tshirt", "color": "white", "pattern": "plain"}, {"category": "tshirt", "color": "white", "pattern": "striped"}, {"category": "tshirt", "color": "white", "pattern": "dotted"}, {"category": "tshirt", "color": "purple", "pattern": "plain"}, {"category": "tshirt", "color": "purple", "pattern": "striped"}, {"category": "tshirt", "color": "purple", "pattern": "dotted"}, {"category": "tshirt", "color": "orange", "pattern": "plain"}, {"category": "tshirt", "color": "orange", "pattern": "striped"}, {"category": "tshirt", "color": "orange", "pattern": "dotted"}, {"category": "mug", "color": "red", "pattern": "plain"}, {"category": "mug", "color": "red", "pattern": "striped"}, {"category": "mug", "color": "red", "pattern": "dotted"}, {"category": "mug", "color": "blue", "pattern": "plain"}, {"category": "mug", "color": "blue", "pattern": "striped"}, {"category": "mug", "color": "blue", "pattern": "dotted"}, {"category": "mug", "color": "green", "pattern": "plain"}, {"category": "mug", "color": "green", "pattern": "striped"}, {"category": "mug", "color": "green", "pattern": "dotted"}, {"category": "mug", "color": "yellow", "pattern": "plain"}, {"category": "mug", "color": "yellow", "pattern": "striped"}, {"category": "mug", "color": "yellow", "pattern": "dotted"}, {"category": "mug", "color": "black", "pattern": "plain"}, {"category": "mug", "color": "black", "pattern": "striped"}, {"category": "mug", "color": "black", "pattern": "dotted"}, {"category": "mug", "color": "white", "pattern": "plain"}, {"category": "mug", "color": "white", "pattern": "striped"}, {"category": "mug", "color": "white", "pattern": "dotted"}, {"category": "mug", "color": "purple", "pattern": "plain"}, {"category": "mug", "color": "purple", "pattern": "striped"}, {"category": "mug", "color": "purple", "pattern": "dotted"}, {"category": "mug", "color": "orange", "pattern": "plain"}, {"category": "mug", "color": "orange", "pattern": "striped"}, {"category": "mug", "color": "orange", "pattern": "dotted"}, {"category": "sneaker", "color": "red", "pattern": "plain"}, {"category": "sneaker", "color": "red", "pattern": "striped"}, {"category": "sneaker", "color": "red", "pattern": "dotted"}, {"category": "sneaker", "color": "blue", "pattern": "plain"}, {"category": "sneaker", "color": "blue", "pattern": "striped"}, {"category": "sneaker", "color": "blue", "pattern": "dotted"}, {"category": "sneaker", "color": "green", "pattern": "plain"}, {"category": "sneaker", "color": "green", "pattern": "striped"}, {"category": "sneaker", "color": "green", "pattern": "dotted"}, {"category": "sneaker", "color": "yellow", "pattern": "plain"}, {"category": "sneaker", "color": "yellow", "pattern": "striped"}, {"category": "sneaker", "color": "yellow", "pattern": "dotted"}, {"category": "sneaker", "color": "black", "pattern": "plain"}, {"category": "sneaker", "color": "black", "pattern": "striped"}, {"category": "sneaker", "color": "black", "pattern": "dotted"}, {"category": "sneaker", "color": "white", "pattern": "plain"}, {"category": "sneaker", "color": "white", "pattern": "striped"}, {"category": "sneaker", "color": "white", "pattern": "dotted"}, {"category": "sneaker", "color": "purple", "pattern": "plain"}, {"category": "sneaker", "color": "purple", "pattern": "striped"}, {"category": "sneaker", "color": "purple", "pattern": "dotted"}, {"category": "sneaker", "color": "orange", "pattern": "plain"}, {"category": "sneaker", "color": "orange", "pattern": "striped"}, {"category": "sneaker", "color": "orange", "pattern": "dotted"}, {"category": "bag", "color": "red", "pattern": "plain"}, {"category": "bag", "color": "red", "pattern": "striped"}, {"category": "bag", "color": "red", "pattern": "dotted"}, {"category": "bag", "color": "blue", "pattern": "plain"}, {"category": "bag", "color": "blue", "pattern": "striped"}, {"category": "bag", "color": "blue", "pattern": "dotted"}, {"category": "bag", "color": "green", "pattern": "plain"}, {"category": "bag", "color": "green", "pattern": "striped"}, {"category": "bag", "color": "green", "pattern": "dotted"}, {"category": "bag", "color": "yellow", "pattern": "plain"}, {"category": "bag", "color": "yellow", "pattern": "striped"}, {"category": "bag", "color": "yellow", "pattern": "dotted"}, {"category": "bag", "color": "black", "pattern": "plain"}, {"category": "bag", "color": "black", "pattern": "striped"}, {"category": "bag", "color": "black", "pattern": "dotted"}, {"category": "bag", "color": "white", "pattern": "plain"}, {"category": "bag", "color": "white", "pattern": "striped"}, {"category": "bag", "color": "white", "pattern": "dotted"}, {"category": "bag", "color": "purple", "pattern": "plain"}, {"category": "bag", "color": "purple", "pattern": "striped"}, {"category": "bag", "color": "purple", "pattern": "dotted"}, {"category": "bag", "color": "orange", "pattern": "plain"}, {"category": "bag", "color": "orange", "pattern": "striped"}, {"category": "bag", "color": "orange", "pattern": "dotted"}, {"category": "hat", "color": "red", "pattern": "plain"}, {"category": "hat", "color": "red", "pattern": "striped"}, {"category": "hat", "color": "red", "pattern": "dotted"}, {"category": "hat", "color": "blue", "pattern": "plain"}, {"category": "hat", "color": "blue", "pattern": "striped"}, {"category": "hat", "color": "blue", "pattern": "dotted"}, {"category": "hat", "color": "green", "pattern": "plain"}, {"category": "hat", "color": "green", "pattern": "striped"}, {"category": "hat", "color": "green", "pattern": "dotted"}, {"category": "hat", "color": "yellow", "pattern": "plain"}, {"category": "hat", "color": "yellow", "pattern": "striped"}, {"category": "hat", "color": "yellow", "pattern": "dotted"}, {"category": "hat", "color": "black", "pattern": "plain"}, {"category": "hat", "color": "black", "pattern": "striped"}, {"category": "hat", "color": "black", "pattern": "dotted"}, {"category": "hat", "color": "white", "pattern": "plain"}, {"category": "hat", "color": "white", "pattern": "striped"}, {"category": "hat", "color": "white", "pattern": "dotted"}, {"category": "hat", "color": "purple", "pattern": "plain"}, {"category": "hat", "color": "purple", "pattern": "striped"}, {"category": "hat", "color": "purple", "pattern": "dotted"}, {"category": "hat", "color": "orange", "pattern": "plain"}, {"category": "hat", "color": "orange", "pattern": "striped"}, {"category": "hat", "color": "orange", "pattern": "dotted"}, {"category": "chair", "color": "red", "pattern": "plain"}, {"category": "chair", "color": "red", "pattern": "striped"}, {"category": "chair", "color": "red", "pattern": "dotted"}, {"category": "chair", "color": "blue", "pattern": "plain"}, {"category": "chair", "color": "blue", "pattern": "striped"}, {"category": "chair", "color": "blue", "pattern": "dotted"}, {"category": "chair", "color": "green", "pattern": "plain"}, {"category": "chair", "color": "green", "pattern": "striped"}, {"category": "chair", "color": "green", "pattern": "dotted"}, {"category": "chair", "color": "yellow", "pattern": "plain"}, {"category": "chair", "color": "yellow", "pattern": "striped"}, {"category": "chair", "color": "yellow", "pattern": "dotted"}, {"category": "chair", "color": "black", "pattern": "plain"}, {"category": "chair", "color": "black", "pattern": "striped"}, {"category": "chair", "color": "black", "pattern": "dotted"}, {"category": "chair", "color": "white", "pattern": "plain"}, {"category": "chair", "color": "white", "pattern": "striped"}, {"category": "chair", "color": "white", "pattern": "dotted"}, {"category": "chair", "color": "purple", "pattern": "plain"}, {"category": "chair", "color": "purple", "pattern": "striped"}, {"category": "chair", "color": "purple", "pattern": "dotted"}, {"category": "chair", "color": "orange", "pattern": "plain"}, {"category": "chair", "color": "orange", "pattern": "striped"}, {"category": "chair", "color": "orange", "pattern": "dotted"}, {"category": "lamp", "color": "red", "pattern": "plain"}, {"category": "lamp", "color": "red", "pattern": "striped"}, {"category": "lamp", "color": "red", "pattern": "dotted"}, {"category": "lamp", "color": "blue", "pattern": "plain"}, {"category": "lamp", "color": "blue", "pattern": "striped"}, {"category": "lamp", "color": "blue", "pattern": "dotted"}, {"category": "lamp", "color": "green", "pattern": "plain"}, {"category": "lamp", "color": "green", "pattern": "striped"}, {"category": "lamp", "color": "green", "pattern": "dotted"}, {"category": "lamp", "color": "yellow", "pattern": "plain"}, {"category": "lamp", "color": "yellow", "pattern": "striped"}, {"category": "lamp", "color": "yellow", "pattern": "dotted"}, {"category": "lamp", "color": "black", "pattern": "plain"}, {"category": "lamp", "color": "black", "pattern": "striped"}, {"category": "lamp", "color": "black", "pattern": "dotted"}, {"category": "lamp", "color": "white", "pattern": "plain"}, {"category": "lamp", "color": "white", "pattern": "striped"}, {"category": "lamp", "color": "white", "pattern": "dotted"}, {"category": "lamp", "color": "purple", "pattern": "plain"}, {"category": "lamp", "color": "purple", "pattern": "striped"}, {"category": "lamp", "color": "purple", "pattern": "dotted"}, {"category": "lamp", "color": "orange", "pattern": "plain"}, {"category": "lamp", "color": "orange", "pattern": "striped"}, {"category": "lamp", "color": "orange", "pattern": "dotted"}, {"category": "bottle", "color": "red", "pattern": "plain"}, {"category": "bottle", "color": "red", "pattern": "striped"}, {"category": "bottle", "color": "red", "pattern": "dotted"}, {"category": "bottle", "color": "blue", "pattern": "plain"}, {"category": "bottle", "color": "blue", "pattern": "striped"}, {"category": "bottle", "color": "blue", "pattern": "dotted"}, {"category": "bottle", "color": "green", "pattern": "plain"}, {"category": "bottle", "color": "green", "pattern": "striped"}, {"category": "bottle", "color": "green", "pattern": "dotted"}, {"category": "bottle", "color": "yellow", "pattern": "plain"}, {"category": "bottle", "color": "yellow", "pattern": "striped"}, {"category": "bottle", "color": "yellow", "pattern": "dotted"}, {"category": "bottle", "color": "black", "pattern": "plain"}, {"category": "bottle", "color": "black", "pattern": "striped"}, {"category": "bottle", "color": "black", "pattern": "dotted"}, {"category": "bottle", "color": "white", "pattern": "plain"}, {"category": "bottle", "color": "white", "pattern": "striped"}, {"category": "bottle", "color": "white", "pattern": "dotted"}, {"category": "bottle", "color": "purple", "pattern": "plain"}, {"category": "bottle", "color": "purple", "pattern": "striped"}, {"category": "bottle", "color": "purple", "pattern": "dotted"}, {"category": "bottle", "color": "orange", "pattern": "plain"}, {"category": "bottle", "color": "orange", "pattern": "striped"}, {"category": "bottle", "color": "orange", "pattern": "dotted"}], "heldout_combos": [["bag", "green", "dotted"], ["bag", "green", "plain"], ["bag", "green", "striped"], ["bag", "orange", "dotted"], ["bag", "white", "striped"], ["bottle", "yellow", "dotted"], ["chair", "black", "plain"], ["chair", "green", "dotted"], ["chair", "green", "plain"], ["hat", "green", "plain"], ["lamp", "black", "plain"], ["lamp", "purple", "dotted"], ["lamp", "red", "plain"], ["lamp", "red", "striped"], ["lamp", "white", "striped"], ["mug", "green", "dotted"], ["mug", "orange", "plain"], ["mug", "yellow", "plain"], ["sneaker", "black", "plain"], ["sneaker", "red", "plain"], ["sneaker", "white", "striped"], ["tshirt", "black", "dotted"], ["tshirt", "orange", "plain"], ["tshirt", "red", "striped"]]}
catalog.png ADDED

Git LFS Details

  • SHA256: 06be47d4ca2c6eae7d0ade0dfa5ebb0a53a8262d233ddf68c017899aa1d78c40
  • Pointer size: 131 Bytes
  • Size of remote file: 788 kB
combined.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d854903eb2fe83b774ca5d1dd66eea40115d822c49a8e6f8d51c8310f889657b
3
+ size 396512
evaluation.json ADDED
@@ -0,0 +1,1045 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "project": "Next Match",
3
+ "profile": "full",
4
+ "seed": 73,
5
+ "device": {
6
+ "python": "3.13.16",
7
+ "platform": "Linux-6.6.159+-x86_64-with-glibc2.39",
8
+ "cpu_count": 12,
9
+ "numpy": "2.4.6",
10
+ "torch": "2.14.1+cu130",
11
+ "cuda": true,
12
+ "gpu": "NVIDIA L4",
13
+ "colab": true
14
+ },
15
+ "data": {
16
+ "type": "synthetic catalog rendered by render.py",
17
+ "combinations": 192,
18
+ "heldout_combinations": [
19
+ [
20
+ "bag",
21
+ "green",
22
+ "dotted"
23
+ ],
24
+ [
25
+ "bag",
26
+ "green",
27
+ "plain"
28
+ ],
29
+ [
30
+ "bag",
31
+ "green",
32
+ "striped"
33
+ ],
34
+ [
35
+ "bag",
36
+ "orange",
37
+ "dotted"
38
+ ],
39
+ [
40
+ "bag",
41
+ "white",
42
+ "striped"
43
+ ],
44
+ [
45
+ "bottle",
46
+ "yellow",
47
+ "dotted"
48
+ ],
49
+ [
50
+ "chair",
51
+ "black",
52
+ "plain"
53
+ ],
54
+ [
55
+ "chair",
56
+ "green",
57
+ "dotted"
58
+ ],
59
+ [
60
+ "chair",
61
+ "green",
62
+ "plain"
63
+ ],
64
+ [
65
+ "hat",
66
+ "green",
67
+ "plain"
68
+ ],
69
+ [
70
+ "lamp",
71
+ "black",
72
+ "plain"
73
+ ],
74
+ [
75
+ "lamp",
76
+ "purple",
77
+ "dotted"
78
+ ],
79
+ [
80
+ "lamp",
81
+ "red",
82
+ "plain"
83
+ ],
84
+ [
85
+ "lamp",
86
+ "red",
87
+ "striped"
88
+ ],
89
+ [
90
+ "lamp",
91
+ "white",
92
+ "striped"
93
+ ],
94
+ [
95
+ "mug",
96
+ "green",
97
+ "dotted"
98
+ ],
99
+ [
100
+ "mug",
101
+ "orange",
102
+ "plain"
103
+ ],
104
+ [
105
+ "mug",
106
+ "yellow",
107
+ "plain"
108
+ ],
109
+ [
110
+ "sneaker",
111
+ "black",
112
+ "plain"
113
+ ],
114
+ [
115
+ "sneaker",
116
+ "red",
117
+ "plain"
118
+ ],
119
+ [
120
+ "sneaker",
121
+ "white",
122
+ "striped"
123
+ ],
124
+ [
125
+ "tshirt",
126
+ "black",
127
+ "dotted"
128
+ ],
129
+ [
130
+ "tshirt",
131
+ "orange",
132
+ "plain"
133
+ ],
134
+ [
135
+ "tshirt",
136
+ "red",
137
+ "striped"
138
+ ]
139
+ ],
140
+ "train_instances_per_combo": 10,
141
+ "gallery": "1 unseen rendering per combination (seed 2)",
142
+ "references": "unseen renderings (seed 3)",
143
+ "templates": {
144
+ "train": {
145
+ "color": {
146
+ "tr": [
147
+ "bunun {c} olanı",
148
+ "{c} renklisi",
149
+ "aynısının {c} olanı",
150
+ "{c} olsun",
151
+ "rengi {c} olsun"
152
+ ],
153
+ "en": [
154
+ "the {c} one",
155
+ "make it {c}",
156
+ "same but {c}",
157
+ "in {c}"
158
+ ]
159
+ },
160
+ "pattern": {
161
+ "tr": [
162
+ "{p} olanı",
163
+ "{p} olsun",
164
+ "aynısı ama {p}"
165
+ ],
166
+ "en": [
167
+ "the {p} one",
168
+ "make it {p}",
169
+ "same but {p}"
170
+ ]
171
+ },
172
+ "category": {
173
+ "tr": [
174
+ "aynısı ama {k}",
175
+ "bunun {k} olanı",
176
+ "aynı desende {k}"
177
+ ],
178
+ "en": [
179
+ "same but a {k}",
180
+ "a {k} like this",
181
+ "same style {k}"
182
+ ]
183
+ },
184
+ "color+pattern": {
185
+ "tr": [
186
+ "{c} ve {p} olanı",
187
+ "{p} ve {c} olsun"
188
+ ],
189
+ "en": [
190
+ "{c} and {p}",
191
+ "make it {p} and {c}"
192
+ ]
193
+ }
194
+ },
195
+ "heldout": {
196
+ "color": {
197
+ "tr": [
198
+ "{c} olanını göster",
199
+ "bundan {c} var mı"
200
+ ],
201
+ "en": [
202
+ "show me the {c} one",
203
+ "do you have it in {c}"
204
+ ]
205
+ },
206
+ "pattern": {
207
+ "tr": [
208
+ "{p} modelini göster"
209
+ ],
210
+ "en": [
211
+ "show a {p} version"
212
+ ]
213
+ },
214
+ "category": {
215
+ "tr": [
216
+ "bunun gibi bir {k} istiyorum"
217
+ ],
218
+ "en": [
219
+ "I want a {k} like this"
220
+ ]
221
+ },
222
+ "color+pattern": {
223
+ "tr": [
224
+ "{c} renkli ve {p} göster"
225
+ ],
226
+ "en": [
227
+ "show it {p} in {c}"
228
+ ]
229
+ }
230
+ }
231
+ },
232
+ "vocabulary_size": 61
233
+ },
234
+ "training": {
235
+ "combined": {
236
+ "seconds": 161.6,
237
+ "final_loss": 0.5543771022558213,
238
+ "steps": 6000,
239
+ "batch": 128
240
+ },
241
+ "image": {
242
+ "seconds": 158.2,
243
+ "final_loss": 3.290284743309021,
244
+ "steps": 6000,
245
+ "batch": 128
246
+ },
247
+ "text": {
248
+ "seconds": 138.7,
249
+ "final_loss": 2.8225105476379393,
250
+ "steps": 6000,
251
+ "batch": 128
252
+ }
253
+ },
254
+ "budget": "same encoder architecture, batch, steps, optimiser and schedule for all three models",
255
+ "results": {
256
+ "all": {
257
+ "n": 1600,
258
+ "combined": {
259
+ "R@1": 0.9325,
260
+ "R@5": 0.96625,
261
+ "R@10": 0.979375
262
+ },
263
+ "image": {
264
+ "R@1": 0.0,
265
+ "R@5": 0.315625,
266
+ "R@10": 0.578125
267
+ },
268
+ "text": {
269
+ "R@1": 0.054375,
270
+ "R@5": 0.263125,
271
+ "R@10": 0.47125
272
+ }
273
+ },
274
+ "edit=color": {
275
+ "n": 648,
276
+ "combined": {
277
+ "R@1": 0.9938271604938271,
278
+ "R@5": 1.0,
279
+ "R@10": 1.0
280
+ },
281
+ "image": {
282
+ "R@1": 0.0,
283
+ "R@5": 0.35802469135802467,
284
+ "R@10": 0.9583333333333334
285
+ },
286
+ "text": {
287
+ "R@1": 0.05092592592592592,
288
+ "R@5": 0.2175925925925926,
289
+ "R@10": 0.39197530864197533
290
+ }
291
+ },
292
+ "edit=pattern": {
293
+ "n": 311,
294
+ "combined": {
295
+ "R@1": 0.8360128617363344,
296
+ "R@5": 0.8971061093247589,
297
+ "R@10": 0.9356913183279743
298
+ },
299
+ "image": {
300
+ "R@1": 0.0,
301
+ "R@5": 0.8778135048231511,
302
+ "R@10": 0.9678456591639871
303
+ },
304
+ "text": {
305
+ "R@1": 0.01607717041800643,
306
+ "R@5": 0.07395498392282958,
307
+ "R@10": 0.19614147909967847
308
+ }
309
+ },
310
+ "edit=category": {
311
+ "n": 322,
312
+ "combined": {
313
+ "R@1": 1.0,
314
+ "R@5": 1.0,
315
+ "R@10": 1.0
316
+ },
317
+ "image": {
318
+ "R@1": 0.0,
319
+ "R@5": 0.0,
320
+ "R@10": 0.0
321
+ },
322
+ "text": {
323
+ "R@1": 0.031055900621118012,
324
+ "R@5": 0.18322981366459629,
325
+ "R@10": 0.37267080745341613
326
+ }
327
+ },
328
+ "edit=color+pattern": {
329
+ "n": 319,
330
+ "combined": {
331
+ "R@1": 0.8338557993730408,
332
+ "R@5": 0.9310344827586207,
333
+ "R@10": 0.9592476489028213
334
+ },
335
+ "image": {
336
+ "R@1": 0.0,
337
+ "R@5": 0.0,
338
+ "R@10": 0.009404388714733543
339
+ },
340
+ "text": {
341
+ "R@1": 0.12225705329153605,
342
+ "R@5": 0.6206896551724138,
343
+ "R@10": 1.0
344
+ }
345
+ },
346
+ "lang=tr": {
347
+ "n": 747,
348
+ "combined": {
349
+ "R@1": 0.9946452476572959,
350
+ "R@5": 1.0,
351
+ "R@10": 1.0
352
+ },
353
+ "image": {
354
+ "R@1": 0.0,
355
+ "R@5": 0.3172690763052209,
356
+ "R@10": 0.570281124497992
357
+ },
358
+ "text": {
359
+ "R@1": 0.050870147255689425,
360
+ "R@5": 0.2583668005354752,
361
+ "R@10": 0.47523427041499333
362
+ }
363
+ },
364
+ "lang=en": {
365
+ "n": 853,
366
+ "combined": {
367
+ "R@1": 0.8780773739742087,
368
+ "R@5": 0.936694021101993,
369
+ "R@10": 0.9613130128956624
370
+ },
371
+ "image": {
372
+ "R@1": 0.0,
373
+ "R@5": 0.3141852286049238,
374
+ "R@10": 0.5849941383352872
375
+ },
376
+ "text": {
377
+ "R@1": 0.05744431418522861,
378
+ "R@5": 0.2672919109026964,
379
+ "R@10": 0.46776084407971863
380
+ }
381
+ },
382
+ "template=seen": {
383
+ "n": 1104,
384
+ "combined": {
385
+ "R@1": 1.0,
386
+ "R@5": 1.0,
387
+ "R@10": 1.0
388
+ },
389
+ "image": {
390
+ "R@1": 0.0,
391
+ "R@5": 0.3233695652173913,
392
+ "R@10": 0.5769927536231884
393
+ },
394
+ "text": {
395
+ "R@1": 0.057065217391304345,
396
+ "R@5": 0.2509057971014493,
397
+ "R@10": 0.45742753623188404
398
+ }
399
+ },
400
+ "template=held-out": {
401
+ "n": 496,
402
+ "combined": {
403
+ "R@1": 0.782258064516129,
404
+ "R@5": 0.8911290322580645,
405
+ "R@10": 0.9334677419354839
406
+ },
407
+ "image": {
408
+ "R@1": 0.0,
409
+ "R@5": 0.29838709677419356,
410
+ "R@10": 0.5806451612903226
411
+ },
412
+ "text": {
413
+ "R@1": 0.04838709677419355,
414
+ "R@5": 0.2903225806451613,
415
+ "R@10": 0.5020161290322581
416
+ }
417
+ },
418
+ "combo=seen": {
419
+ "n": 1192,
420
+ "combined": {
421
+ "R@1": 0.9328859060402684,
422
+ "R@5": 0.9664429530201343,
423
+ "R@10": 0.9773489932885906
424
+ },
425
+ "image": {
426
+ "R@1": 0.0,
427
+ "R@5": 0.34815436241610737,
428
+ "R@10": 0.5956375838926175
429
+ },
430
+ "text": {
431
+ "R@1": 0.05704697986577181,
432
+ "R@5": 0.2961409395973154,
433
+ "R@10": 0.5016778523489933
434
+ }
435
+ },
436
+ "combo=held-out": {
437
+ "n": 408,
438
+ "combined": {
439
+ "R@1": 0.9313725490196079,
440
+ "R@5": 0.9656862745098039,
441
+ "R@10": 0.9852941176470589
442
+ },
443
+ "image": {
444
+ "R@1": 0.0,
445
+ "R@5": 0.22058823529411764,
446
+ "R@10": 0.5269607843137255
447
+ },
448
+ "text": {
449
+ "R@1": 0.04656862745098039,
450
+ "R@5": 0.16666666666666666,
451
+ "R@10": 0.38235294117647056
452
+ }
453
+ }
454
+ },
455
+ "failure_examples_combined": [
456
+ {
457
+ "ref": [
458
+ "lamp",
459
+ "black",
460
+ "striped"
461
+ ],
462
+ "target": [
463
+ "lamp",
464
+ "black",
465
+ "plain"
466
+ ],
467
+ "text": "show a plain version",
468
+ "kind": "pattern",
469
+ "lang": "en",
470
+ "heldout_template": true,
471
+ "unseen_combo": true,
472
+ "rank_combined": 15
473
+ },
474
+ {
475
+ "ref": [
476
+ "mug",
477
+ "blue",
478
+ "dotted"
479
+ ],
480
+ "target": [
481
+ "mug",
482
+ "blue",
483
+ "striped"
484
+ ],
485
+ "text": "show a striped version",
486
+ "kind": "pattern",
487
+ "lang": "en",
488
+ "heldout_template": true,
489
+ "unseen_combo": false,
490
+ "rank_combined": 53
491
+ },
492
+ {
493
+ "ref": [
494
+ "hat",
495
+ "yellow",
496
+ "dotted"
497
+ ],
498
+ "target": [
499
+ "hat",
500
+ "black",
501
+ "striped"
502
+ ],
503
+ "text": "show it striped in black",
504
+ "kind": "color+pattern",
505
+ "lang": "en",
506
+ "heldout_template": true,
507
+ "unseen_combo": false,
508
+ "rank_combined": 16
509
+ },
510
+ {
511
+ "ref": [
512
+ "mug",
513
+ "blue",
514
+ "dotted"
515
+ ],
516
+ "target": [
517
+ "mug",
518
+ "blue",
519
+ "striped"
520
+ ],
521
+ "text": "show a striped version",
522
+ "kind": "pattern",
523
+ "lang": "en",
524
+ "heldout_template": true,
525
+ "unseen_combo": false,
526
+ "rank_combined": 53
527
+ },
528
+ {
529
+ "ref": [
530
+ "bag",
531
+ "black",
532
+ "dotted"
533
+ ],
534
+ "target": [
535
+ "bag",
536
+ "orange",
537
+ "striped"
538
+ ],
539
+ "text": "show it striped in orange",
540
+ "kind": "color+pattern",
541
+ "lang": "en",
542
+ "heldout_template": true,
543
+ "unseen_combo": false,
544
+ "rank_combined": 8
545
+ },
546
+ {
547
+ "ref": [
548
+ "bottle",
549
+ "green",
550
+ "plain"
551
+ ],
552
+ "target": [
553
+ "bottle",
554
+ "white",
555
+ "striped"
556
+ ],
557
+ "text": "show it striped in white",
558
+ "kind": "color+pattern",
559
+ "lang": "en",
560
+ "heldout_template": true,
561
+ "unseen_combo": false,
562
+ "rank_combined": 50
563
+ },
564
+ {
565
+ "ref": [
566
+ "bag",
567
+ "red",
568
+ "plain"
569
+ ],
570
+ "target": [
571
+ "bag",
572
+ "red",
573
+ "striped"
574
+ ],
575
+ "text": "show a striped version",
576
+ "kind": "pattern",
577
+ "lang": "en",
578
+ "heldout_template": true,
579
+ "unseen_combo": false,
580
+ "rank_combined": 64
581
+ },
582
+ {
583
+ "ref": [
584
+ "hat",
585
+ "purple",
586
+ "striped"
587
+ ],
588
+ "target": [
589
+ "hat",
590
+ "purple",
591
+ "plain"
592
+ ],
593
+ "text": "show a plain version",
594
+ "kind": "pattern",
595
+ "lang": "en",
596
+ "heldout_template": true,
597
+ "unseen_combo": false,
598
+ "rank_combined": 25
599
+ },
600
+ {
601
+ "ref": [
602
+ "bottle",
603
+ "green",
604
+ "plain"
605
+ ],
606
+ "target": [
607
+ "bottle",
608
+ "white",
609
+ "striped"
610
+ ],
611
+ "text": "show it striped in white",
612
+ "kind": "color+pattern",
613
+ "lang": "en",
614
+ "heldout_template": true,
615
+ "unseen_combo": false,
616
+ "rank_combined": 50
617
+ },
618
+ {
619
+ "ref": [
620
+ "bag",
621
+ "purple",
622
+ "dotted"
623
+ ],
624
+ "target": [
625
+ "bag",
626
+ "green",
627
+ "striped"
628
+ ],
629
+ "text": "show it striped in green",
630
+ "kind": "color+pattern",
631
+ "lang": "en",
632
+ "heldout_template": true,
633
+ "unseen_combo": true,
634
+ "rank_combined": 6
635
+ },
636
+ {
637
+ "ref": [
638
+ "mug",
639
+ "red",
640
+ "dotted"
641
+ ],
642
+ "target": [
643
+ "mug",
644
+ "red",
645
+ "striped"
646
+ ],
647
+ "text": "show a striped version",
648
+ "kind": "pattern",
649
+ "lang": "en",
650
+ "heldout_template": true,
651
+ "unseen_combo": false,
652
+ "rank_combined": 28
653
+ },
654
+ {
655
+ "ref": [
656
+ "sneaker",
657
+ "white",
658
+ "plain"
659
+ ],
660
+ "target": [
661
+ "sneaker",
662
+ "white",
663
+ "striped"
664
+ ],
665
+ "text": "show a striped version",
666
+ "kind": "pattern",
667
+ "lang": "en",
668
+ "heldout_template": true,
669
+ "unseen_combo": true,
670
+ "rank_combined": 60
671
+ },
672
+ {
673
+ "ref": [
674
+ "hat",
675
+ "red",
676
+ "dotted"
677
+ ],
678
+ "target": [
679
+ "hat",
680
+ "black",
681
+ "striped"
682
+ ],
683
+ "text": "show it striped in black",
684
+ "kind": "color+pattern",
685
+ "lang": "en",
686
+ "heldout_template": true,
687
+ "unseen_combo": false,
688
+ "rank_combined": 5
689
+ },
690
+ {
691
+ "ref": [
692
+ "bottle",
693
+ "purple",
694
+ "dotted"
695
+ ],
696
+ "target": [
697
+ "bottle",
698
+ "purple",
699
+ "striped"
700
+ ],
701
+ "text": "show a striped version",
702
+ "kind": "pattern",
703
+ "lang": "en",
704
+ "heldout_template": true,
705
+ "unseen_combo": false,
706
+ "rank_combined": 8
707
+ },
708
+ {
709
+ "ref": [
710
+ "bag",
711
+ "blue",
712
+ "dotted"
713
+ ],
714
+ "target": [
715
+ "bag",
716
+ "blue",
717
+ "plain"
718
+ ],
719
+ "text": "show a plain version",
720
+ "kind": "pattern",
721
+ "lang": "en",
722
+ "heldout_template": true,
723
+ "unseen_combo": false,
724
+ "rank_combined": 22
725
+ },
726
+ {
727
+ "ref": [
728
+ "sneaker",
729
+ "red",
730
+ "striped"
731
+ ],
732
+ "target": [
733
+ "sneaker",
734
+ "red",
735
+ "plain"
736
+ ],
737
+ "text": "show a plain version",
738
+ "kind": "pattern",
739
+ "lang": "en",
740
+ "heldout_template": true,
741
+ "unseen_combo": true,
742
+ "rank_combined": 6
743
+ },
744
+ {
745
+ "ref": [
746
+ "mug",
747
+ "red",
748
+ "striped"
749
+ ],
750
+ "target": [
751
+ "mug",
752
+ "red",
753
+ "plain"
754
+ ],
755
+ "text": "show a plain version",
756
+ "kind": "pattern",
757
+ "lang": "en",
758
+ "heldout_template": true,
759
+ "unseen_combo": false,
760
+ "rank_combined": 73
761
+ },
762
+ {
763
+ "ref": [
764
+ "mug",
765
+ "purple",
766
+ "dotted"
767
+ ],
768
+ "target": [
769
+ "mug",
770
+ "purple",
771
+ "plain"
772
+ ],
773
+ "text": "show a plain version",
774
+ "kind": "pattern",
775
+ "lang": "en",
776
+ "heldout_template": true,
777
+ "unseen_combo": false,
778
+ "rank_combined": 125
779
+ },
780
+ {
781
+ "ref": [
782
+ "mug",
783
+ "red",
784
+ "dotted"
785
+ ],
786
+ "target": [
787
+ "mug",
788
+ "red",
789
+ "plain"
790
+ ],
791
+ "text": "show a plain version",
792
+ "kind": "pattern",
793
+ "lang": "en",
794
+ "heldout_template": true,
795
+ "unseen_combo": false,
796
+ "rank_combined": 89
797
+ },
798
+ {
799
+ "ref": [
800
+ "sneaker",
801
+ "green",
802
+ "striped"
803
+ ],
804
+ "target": [
805
+ "sneaker",
806
+ "green",
807
+ "dotted"
808
+ ],
809
+ "text": "show a dotted version",
810
+ "kind": "pattern",
811
+ "lang": "en",
812
+ "heldout_template": true,
813
+ "unseen_combo": false,
814
+ "rank_combined": 8
815
+ },
816
+ {
817
+ "ref": [
818
+ "lamp",
819
+ "red",
820
+ "dotted"
821
+ ],
822
+ "target": [
823
+ "lamp",
824
+ "black",
825
+ "striped"
826
+ ],
827
+ "text": "show it striped in black",
828
+ "kind": "color+pattern",
829
+ "lang": "en",
830
+ "heldout_template": true,
831
+ "unseen_combo": false,
832
+ "rank_combined": 5
833
+ },
834
+ {
835
+ "ref": [
836
+ "tshirt",
837
+ "white",
838
+ "plain"
839
+ ],
840
+ "target": [
841
+ "tshirt",
842
+ "white",
843
+ "striped"
844
+ ],
845
+ "text": "show a striped version",
846
+ "kind": "pattern",
847
+ "lang": "en",
848
+ "heldout_template": true,
849
+ "unseen_combo": false,
850
+ "rank_combined": 15
851
+ },
852
+ {
853
+ "ref": [
854
+ "sneaker",
855
+ "black",
856
+ "dotted"
857
+ ],
858
+ "target": [
859
+ "sneaker",
860
+ "black",
861
+ "striped"
862
+ ],
863
+ "text": "show a striped version",
864
+ "kind": "pattern",
865
+ "lang": "en",
866
+ "heldout_template": true,
867
+ "unseen_combo": false,
868
+ "rank_combined": 80
869
+ },
870
+ {
871
+ "ref": [
872
+ "bottle",
873
+ "black",
874
+ "plain"
875
+ ],
876
+ "target": [
877
+ "bottle",
878
+ "blue",
879
+ "striped"
880
+ ],
881
+ "text": "show it striped in blue",
882
+ "kind": "color+pattern",
883
+ "lang": "en",
884
+ "heldout_template": true,
885
+ "unseen_combo": false,
886
+ "rank_combined": 23
887
+ },
888
+ {
889
+ "ref": [
890
+ "chair",
891
+ "black",
892
+ "striped"
893
+ ],
894
+ "target": [
895
+ "chair",
896
+ "black",
897
+ "plain"
898
+ ],
899
+ "text": "show a plain version",
900
+ "kind": "pattern",
901
+ "lang": "en",
902
+ "heldout_template": true,
903
+ "unseen_combo": true,
904
+ "rank_combined": 6
905
+ },
906
+ {
907
+ "ref": [
908
+ "sneaker",
909
+ "blue",
910
+ "plain"
911
+ ],
912
+ "target": [
913
+ "sneaker",
914
+ "blue",
915
+ "striped"
916
+ ],
917
+ "text": "show a striped version",
918
+ "kind": "pattern",
919
+ "lang": "en",
920
+ "heldout_template": true,
921
+ "unseen_combo": false,
922
+ "rank_combined": 57
923
+ },
924
+ {
925
+ "ref": [
926
+ "chair",
927
+ "orange",
928
+ "striped"
929
+ ],
930
+ "target": [
931
+ "chair",
932
+ "orange",
933
+ "plain"
934
+ ],
935
+ "text": "show a plain version",
936
+ "kind": "pattern",
937
+ "lang": "en",
938
+ "heldout_template": true,
939
+ "unseen_combo": false,
940
+ "rank_combined": 10
941
+ },
942
+ {
943
+ "ref": [
944
+ "bag",
945
+ "green",
946
+ "dotted"
947
+ ],
948
+ "target": [
949
+ "bag",
950
+ "green",
951
+ "plain"
952
+ ],
953
+ "text": "show a plain version",
954
+ "kind": "pattern",
955
+ "lang": "en",
956
+ "heldout_template": true,
957
+ "unseen_combo": true,
958
+ "rank_combined": 11
959
+ },
960
+ {
961
+ "ref": [
962
+ "sneaker",
963
+ "yellow",
964
+ "dotted"
965
+ ],
966
+ "target": [
967
+ "sneaker",
968
+ "orange",
969
+ "striped"
970
+ ],
971
+ "text": "show it striped in orange",
972
+ "kind": "color+pattern",
973
+ "lang": "en",
974
+ "heldout_template": true,
975
+ "unseen_combo": false,
976
+ "rank_combined": 24
977
+ },
978
+ {
979
+ "ref": [
980
+ "hat",
981
+ "orange",
982
+ "dotted"
983
+ ],
984
+ "target": [
985
+ "hat",
986
+ "white",
987
+ "striped"
988
+ ],
989
+ "text": "show it striped in white",
990
+ "kind": "color+pattern",
991
+ "lang": "en",
992
+ "heldout_template": true,
993
+ "unseen_combo": false,
994
+ "rank_combined": 12
995
+ }
996
+ ],
997
+ "tests": {
998
+ "reload": {
999
+ "returncode": 0,
1000
+ "seconds": 0.24,
1001
+ "process": "fresh python -E subprocess, cwd = temporary copy of the model folder",
1002
+ "passed": true,
1003
+ "cases": 6,
1004
+ "mismatches": []
1005
+ },
1006
+ "node_parity": {
1007
+ "cases": 6,
1008
+ "failed": 0,
1009
+ "maxDiff": 1.6653345369377348e-16,
1010
+ "results": [
1011
+ {
1012
+ "name": "28:bunun kırmızı olanı",
1013
+ "maxDiff": 1.3877787807814457e-17
1014
+ },
1015
+ {
1016
+ "name": "17:çizgili olanı",
1017
+ "maxDiff": 5.551115123125783e-17
1018
+ },
1019
+ {
1020
+ "name": "102:same but a lamp",
1021
+ "maxDiff": 2.7755575615628914e-17
1022
+ },
1023
+ {
1024
+ "name": "84:make it dotted and yellow",
1025
+ "maxDiff": 1.249000902703301e-16
1026
+ },
1027
+ {
1028
+ "name": "190:İSTİYORUM, mor!",
1029
+ "maxDiff": 2.949029909160572e-17
1030
+ },
1031
+ {
1032
+ "name": "122:",
1033
+ "maxDiff": 1.6653345369377348e-16
1034
+ }
1035
+ ],
1036
+ "ms": 749,
1037
+ "runtime": "node v22.23.4",
1038
+ "status": "passed"
1039
+ },
1040
+ "browser": {
1041
+ "status": "skipped",
1042
+ "reason": "playwright not installed"
1043
+ }
1044
+ }
1045
+ }
image.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78938b999126b7e88a9a90d8aebefc7099fba62a351be66f6876a67b222e4460
3
+ size 297432
inference.py ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Next Match inference (NumPy only): rank catalog items for a reference image plus a text edit.
2
+
3
+ Usage:
4
+ python inference.py --selftest
5
+ from inference import load, search
6
+ ids = search(spec, ref_image_uint8_hwc, "bunun kırmızı olanı")
7
+ """
8
+ import json
9
+ import os
10
+ import sys
11
+
12
+ HERE = os.path.dirname(os.path.abspath(__file__))
13
+ sys.path.insert(0, HERE)
14
+
15
+ import numpy as np # noqa: E402
16
+
17
+ import match_core as mc # noqa: E402
18
+ from nlrt import load_model # noqa: E402
19
+
20
+
21
+ def load(path=os.path.join(HERE, "model.json")):
22
+ return load_model(path)
23
+
24
+
25
+ def to_chw(img_uint8_hwc):
26
+ return np.asarray(img_uint8_hwc, dtype=np.float64).transpose(2, 0, 1) / 255.0
27
+
28
+
29
+ def search(spec, ref_image, text, mode="combined", top=10, keys=None):
30
+ """ref_image: (48, 48, 3) uint8. Returns catalog indices (into the released catalog) best first."""
31
+ q = mc.encode_query(spec, to_chw(ref_image), text, mode)
32
+ name = {"combined": "gallery_keys", "image": "gallery_keys_img_only", "text": "gallery_keys_txt_only"}[mode]
33
+ K = spec["tensors"][name] if keys is None else keys
34
+ s = K @ q
35
+ return sorted(range(len(s)), key=lambda i: (-s[i], i))[:top]
36
+
37
+
38
+ def _selftest():
39
+ spec = load()
40
+ with open(os.path.join(HERE, "selftest.json"), encoding="utf-8") as f:
41
+ cases = json.load(f)["cases"]
42
+ bad = []
43
+ for c in cases:
44
+ img = np.asarray(c["pixels"], dtype=np.uint8).reshape(3, 48, 48).transpose(1, 2, 0)
45
+ got = {m: search(spec, img, c["text"], m, 5) for m in ("combined", "image", "text")}
46
+ if got != c["expected"]:
47
+ bad.append({"text": c["text"], "got": got, "expected": c["expected"]})
48
+ print(json.dumps({"passed": not bad, "cases": len(cases), "mismatches": bad[:2]}, ensure_ascii=False))
49
+ return not bad
50
+
51
+
52
+ if __name__ == "__main__":
53
+ if "--selftest" in sys.argv:
54
+ sys.exit(0 if _selftest() else 1)
55
+ print(__doc__)
match_core.py ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Next Match: search a catalog with a reference image plus a text edit ("bunun kirmizi olani", "same but striped").
2
+
3
+ Shared by the notebook, inference.py and (ported) web/core.js:
4
+ * tokenize: Turkish-aware lower-casing, punctuation removal, longest-prefix stem matching against a fixed vocabulary
5
+ * encode_query: image features + text features -> fused, L2-normalised query vector
6
+ * encode_item: catalog image -> L2-normalised key vector
7
+ The catalog is synthetic: procedurally drawn product icons (8 categories x 8 colours x 3 patterns), rendered in the notebook.
8
+ """
9
+ import numpy as np
10
+
11
+ from nlrt import forward
12
+
13
+ CATEGORIES = ["tshirt", "mug", "sneaker", "bag", "hat", "chair", "lamp", "bottle"]
14
+ COLORS = ["red", "blue", "green", "yellow", "black", "white", "purple", "orange"]
15
+ PATTERNS = ["plain", "striped", "dotted"]
16
+ RGB = {"red": (214, 40, 40), "blue": (40, 90, 210), "green": (40, 160, 70), "yellow": (240, 200, 30),
17
+ "black": (30, 30, 34), "white": (250, 250, 250), "purple": (130, 60, 170), "orange": (245, 130, 30)}
18
+ TR = {"red": "kırmızı", "blue": "mavi", "green": "yeşil", "yellow": "sarı", "black": "siyah", "white": "beyaz", "purple": "mor",
19
+ "orange": "turuncu", "plain": "düz", "striped": "çizgili", "dotted": "puantiyeli", "tshirt": "tişört", "mug": "kupa",
20
+ "sneaker": "ayakkabı", "bag": "çanta", "hat": "şapka", "chair": "sandalye", "lamp": "lamba", "bottle": "şişe"}
21
+ EN = {"tshirt": "t-shirt", "mug": "mug", "sneaker": "sneaker", "bag": "bag", "hat": "hat", "chair": "chair", "lamp": "lamp", "bottle": "bottle",
22
+ "plain": "plain", "striped": "striped", "dotted": "dotted"}
23
+ SYNONYMS = {"tshirt": ["tshirt", "shirt", "tee"], "dotted": ["dotted", "dots", "polka"], "striped": ["striped", "stripes", "stripe"],
24
+ "sneaker": ["sneaker", "shoe", "shoes"], "plain": ["plain", "solid"]}
25
+
26
+
27
+ def lower_tr(s):
28
+ return s.replace("I", "ı").replace("İ", "i").lower()
29
+
30
+
31
+ def tokenize(text, vocab):
32
+ """Return token ids (unknown words dropped). Matches whole words, then the longest vocabulary prefix of >= 4 letters."""
33
+ clean = "".join(ch if (ch.isalnum() or ch == " ") else " " for ch in lower_tr(text).replace("-", ""))
34
+ ids = []
35
+ for w in clean.split():
36
+ if w in vocab:
37
+ ids.append(vocab[w])
38
+ continue
39
+ best = None
40
+ for k in range(len(w) - 1, 3, -1):
41
+ if w[:k] in vocab:
42
+ best = vocab[w[:k]]
43
+ break
44
+ if best is not None:
45
+ ids.append(best)
46
+ return ids
47
+
48
+
49
+ def text_vector(ids, emb):
50
+ if not ids:
51
+ return np.zeros(emb.shape[1])
52
+ return np.mean(emb[ids], axis=0)
53
+
54
+
55
+ def l2n(x):
56
+ return x / max(float(np.linalg.norm(x)), 1e-12)
57
+
58
+
59
+ def image_features(spec, images, net="img"):
60
+ """images: (N, 3, 48, 48) floats in [0, 1]."""
61
+ return forward(spec["nets"][net], np.asarray(images, dtype=np.float64))
62
+
63
+
64
+ def encode_item(spec, images, net="img"):
65
+ f = image_features(spec, images, net)
66
+ return f / np.maximum(np.linalg.norm(f, axis=1, keepdims=True), 1e-12)
67
+
68
+
69
+ def encode_query(spec, image, text, mode="combined"):
70
+ """mode: combined | image | text. image: (3, 48, 48) in [0, 1]."""
71
+ vocab = spec["meta"]["vocab"]
72
+ if mode == "combined":
73
+ fi = image_features(spec, image[None], "img")[0]
74
+ ft = forward(spec["nets"]["txt"], text_vector(tokenize(text, vocab), spec["tensors"]["emb"])[None])[0]
75
+ return l2n(fi + forward(spec["nets"]["fuse"], np.concatenate([fi, ft])[None])[0])
76
+ if mode == "image":
77
+ return l2n(forward(spec["nets"]["img_only_head"], image_features(spec, image[None], "img_only")[0][None])[0])
78
+ ft = forward(spec["nets"]["txt_only"], text_vector(tokenize(text, vocab), spec["tensors"]["emb_only"])[None])[0]
79
+ return l2n(ft)
model.json ADDED
The diff for this file is too large to render. See raw diff
 
nlrt.py ADDED
@@ -0,0 +1,226 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """nlrt: tiny portable runtime shared by Next Lab models.
2
+
3
+ The same layer list runs in NumPy (this file) and in the browser (nlrt.js).
4
+ Weights are stored as base64 little-endian float32 inside model.json.
5
+ Only NumPy is required to load and run a model.
6
+ """
7
+ import base64
8
+ import json
9
+ import math
10
+
11
+ import numpy as np
12
+
13
+ __all__ = ["b64e", "b64d", "forward", "load_model", "Mulberry32", "export_sequential"]
14
+
15
+
16
+ def b64e(a):
17
+ return base64.b64encode(np.ascontiguousarray(a, dtype="<f4").tobytes()).decode("ascii")
18
+
19
+
20
+ def b64d(s, shape=None):
21
+ a = np.frombuffer(base64.b64decode(s), dtype="<f4").astype(np.float64)
22
+ return a.reshape(shape) if shape is not None else a
23
+
24
+
25
+ def load_model(path):
26
+ with open(path, "r", encoding="utf-8") as f:
27
+ spec = json.load(f)
28
+ for net in spec.get("nets", {}).values():
29
+ _decode(net)
30
+ for k, v in list(spec.get("tensors", {}).items()):
31
+ spec["tensors"][k] = b64d(v["data"], v["shape"])
32
+ return spec
33
+
34
+
35
+ def _decode(layers):
36
+ for L in layers:
37
+ for key in ("w", "b"):
38
+ if key in L and isinstance(L[key], str):
39
+ L[key] = b64d(L[key], L.get(key + "_shape"))
40
+
41
+
42
+ class Mulberry32:
43
+ """Deterministic PRNG with an identical JavaScript twin in nlrt.js."""
44
+
45
+ def __init__(self, seed):
46
+ self.a = seed & 0xFFFFFFFF
47
+
48
+ def random(self):
49
+ self.a = (self.a + 0x6D2B79F5) & 0xFFFFFFFF
50
+ t = self.a
51
+ t = ((t ^ (t >> 15)) * (t | 1)) & 0xFFFFFFFF
52
+ t ^= (t + (((t ^ (t >> 7)) * (t | 61)) & 0xFFFFFFFF)) & 0xFFFFFFFF
53
+ return ((t ^ (t >> 14)) & 0xFFFFFFFF) / 4294967296.0
54
+
55
+ def uniform(self, lo, hi):
56
+ return lo + (hi - lo) * self.random()
57
+
58
+ def randint(self, n):
59
+ return int(math.floor(self.random() * n))
60
+
61
+ def normal(self):
62
+ u = max(self.random(), 1e-12)
63
+ v = self.random()
64
+ return math.sqrt(-2.0 * math.log(u)) * math.cos(2.0 * math.pi * v)
65
+
66
+
67
+ def _pair(v):
68
+ return list(v) if isinstance(v, (list, tuple)) else [v, v]
69
+
70
+
71
+ def _conv2d(x, w, b, stride, pad):
72
+ n, c, h, wd = x.shape
73
+ o, _, kh, kw = w.shape
74
+ sh, sw = stride
75
+ ph, pw = pad
76
+ xp = np.pad(x, ((0, 0), (0, 0), (ph, ph), (pw, pw)))
77
+ oh = (h + 2 * ph - kh) // sh + 1
78
+ ow = (wd + 2 * pw - kw) // sw + 1
79
+ cols = np.empty((n, c, kh, kw, oh, ow), dtype=x.dtype)
80
+ for i in range(kh):
81
+ for j in range(kw):
82
+ cols[:, :, i, j] = xp[:, :, i:i + sh * oh:sh, j:j + sw * ow:sw]
83
+ out = np.tensordot(cols, w, axes=([1, 2, 3], [1, 2, 3])) # n, oh, ow, o
84
+ out = out.transpose(0, 3, 1, 2)
85
+ if b is not None:
86
+ out = out + b[None, :, None, None]
87
+ return out
88
+
89
+
90
+ def _maxpool(x, k):
91
+ kh, kw = k
92
+ n, c, h, w = x.shape
93
+ oh, ow = h // kh, w // kw
94
+ x = x[:, :, :oh * kh, :ow * kw].reshape(n, c, oh, kh, ow, kw)
95
+ return x.max(axis=(3, 5))
96
+
97
+
98
+ def forward(layers, x):
99
+ """Run a layer list on a batch. x: (N, D) or (N, C, H, W) or (N, C, L)."""
100
+ x = np.asarray(x, dtype=np.float64)
101
+ for L in layers:
102
+ t = L["type"]
103
+ if t == "linear":
104
+ x = x @ L["w"].T + L["b"]
105
+ elif t == "conv2d":
106
+ squeeze = x.ndim == 3
107
+ if squeeze:
108
+ x = x[:, :, None, :]
109
+ x = _conv2d(x, L["w"], L["b"], _pair(L["stride"]), _pair(L["pad"]))
110
+ if squeeze:
111
+ x = x[:, :, 0, :]
112
+ elif t == "relu":
113
+ x = np.maximum(x, 0)
114
+ elif t == "tanh":
115
+ x = np.tanh(x)
116
+ elif t == "sigmoid":
117
+ x = 1 / (1 + np.exp(-x))
118
+ elif t == "maxpool":
119
+ squeeze = x.ndim == 3
120
+ if squeeze:
121
+ x = x[:, :, None, :]
122
+ x = _maxpool(x, _pair(L["k"]))
123
+ if squeeze:
124
+ x = x[:, :, 0, :]
125
+ elif t == "gap":
126
+ x = x.reshape(x.shape[0], x.shape[1], -1).mean(axis=2)
127
+ elif t == "flatten":
128
+ x = x.reshape(x.shape[0], -1)
129
+ elif t == "upsample":
130
+ x = x.repeat(L["s"], axis=-1)
131
+ if x.ndim == 4:
132
+ x = x.repeat(L["s"], axis=-2)
133
+ elif t == "l2norm":
134
+ x = x / np.maximum(np.linalg.norm(x, axis=1, keepdims=True), 1e-12)
135
+ elif t == "reshape":
136
+ x = x.reshape([x.shape[0]] + list(L["shape"]))
137
+ else:
138
+ raise ValueError("unknown layer " + t)
139
+ return x
140
+
141
+
142
+ def export_sequential(seq):
143
+ """Convert a torch nn.Sequential into a portable layer list (BatchNorm folded)."""
144
+ import torch.nn as nn
145
+
146
+ mods = list(seq.modules())[1:] if not isinstance(seq, (list, tuple)) else list(seq)
147
+ mods = [m for m in mods if not isinstance(m, nn.Sequential)]
148
+ layers = []
149
+
150
+ def tensor(t):
151
+ return t.detach().cpu().double().numpy()
152
+
153
+ for m in mods:
154
+ if isinstance(m, nn.Linear):
155
+ b = tensor(m.bias) if m.bias is not None else np.zeros(m.out_features)
156
+ layers.append({"type": "linear", "w": tensor(m.weight), "b": b})
157
+ elif isinstance(m, (nn.Conv2d, nn.Conv1d)):
158
+ w = tensor(m.weight)
159
+ b = tensor(m.bias) if m.bias is not None else np.zeros(w.shape[0])
160
+ if isinstance(m, nn.Conv1d):
161
+ w = w[:, :, None, :]
162
+ stride, pad = [1, m.stride[0]], [0, m.padding[0]]
163
+ else:
164
+ stride, pad = list(m.stride), list(m.padding)
165
+ layers.append({"type": "conv2d", "w": w, "b": b, "stride": stride, "pad": pad})
166
+ elif isinstance(m, (nn.BatchNorm1d, nn.BatchNorm2d)):
167
+ prev = layers[-1]
168
+ assert prev["type"] in ("conv2d", "linear"), "BatchNorm must follow conv/linear"
169
+ scale = tensor(m.weight) / np.sqrt(tensor(m.running_var) + m.eps)
170
+ shift = tensor(m.bias) - tensor(m.running_mean) * scale
171
+ if prev["type"] == "conv2d":
172
+ prev["w"] = prev["w"] * scale[:, None, None, None]
173
+ else:
174
+ prev["w"] = prev["w"] * scale[:, None]
175
+ prev["b"] = prev["b"] * scale + shift
176
+ elif isinstance(m, nn.ReLU):
177
+ layers.append({"type": "relu"})
178
+ elif isinstance(m, nn.Tanh):
179
+ layers.append({"type": "tanh"})
180
+ elif isinstance(m, nn.Sigmoid):
181
+ layers.append({"type": "sigmoid"})
182
+ elif isinstance(m, nn.MaxPool2d):
183
+ layers.append({"type": "maxpool", "k": _pair(m.kernel_size)})
184
+ elif isinstance(m, nn.MaxPool1d):
185
+ k = m.kernel_size if isinstance(m, int) else m.kernel_size
186
+ layers.append({"type": "maxpool", "k": [1, k if isinstance(k, int) else k[0]]})
187
+ elif isinstance(m, (nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d)):
188
+ layers.append({"type": "gap"})
189
+ elif isinstance(m, nn.Flatten):
190
+ layers.append({"type": "flatten"})
191
+ elif isinstance(m, nn.Upsample):
192
+ layers.append({"type": "upsample", "s": int(m.scale_factor)})
193
+ elif isinstance(m, nn.Dropout):
194
+ continue
195
+ elif m.__class__.__name__ == "L2Norm":
196
+ layers.append({"type": "l2norm"})
197
+ elif m.__class__.__name__ == "Reshape":
198
+ layers.append({"type": "reshape", "shape": list(m.shape)})
199
+ else:
200
+ raise ValueError("cannot export " + m.__class__.__name__)
201
+ return layers
202
+
203
+
204
+ def encode_layers(layers):
205
+ out = []
206
+ for L in layers:
207
+ L2 = dict(L)
208
+ for key in ("w", "b"):
209
+ if key in L2:
210
+ arr = np.asarray(L2[key])
211
+ L2[key + "_shape"] = list(arr.shape)
212
+ L2[key] = b64e(arr)
213
+ out.append(L2)
214
+ return out
215
+
216
+
217
+ def save_model(path, nets=None, tensors=None, meta=None):
218
+ spec = {"format": "nlrt-1", "meta": meta or {}, "nets": {}, "tensors": {}}
219
+ for name, layers in (nets or {}).items():
220
+ spec["nets"][name] = encode_layers(layers)
221
+ for name, arr in (tensors or {}).items():
222
+ arr = np.asarray(arr)
223
+ spec["tensors"][name] = {"shape": list(arr.shape), "data": b64e(arr)}
224
+ with open(path, "w", encoding="utf-8") as f:
225
+ json.dump(spec, f, separators=(",", ":"))
226
+ return spec
preview.png ADDED

Git LFS Details

  • SHA256: 502806df91e0eb55e17bbd33e4b0b95632b35434fcf8ad4fc01fd8f386bda051
  • Pointer size: 131 Bytes
  • Size of remote file: 377 kB
render.py ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Procedural product icons for the synthetic Next Match catalog (48x48 RGB, uint8)."""
2
+ import numpy as np
3
+ from PIL import Image, ImageDraw
4
+
5
+ from match_core import RGB
6
+
7
+ S = 48
8
+
9
+
10
+ def _shape(draw, cat, k, ox, oy):
11
+ """Draw category silhouette (fill=255) into an L-mode mask. k = scale, (ox, oy) = offset."""
12
+ P = lambda pts: [(ox + k * x, oy + k * y) for x, y in pts] # noqa: E731
13
+ if cat == "tshirt":
14
+ draw.polygon(P([(8, 10), (17, 6), (31, 6), (40, 10), (44, 19), (36, 22), (35, 42), (13, 42), (12, 22), (4, 19)]), fill=255)
15
+ elif cat == "mug":
16
+ draw.rectangle(P([(10, 12), (32, 42)]), fill=255)
17
+ draw.ellipse(P([(26, 18), (42, 34)]), fill=255)
18
+ draw.ellipse(P([(30, 22), (38, 30)]), fill=0)
19
+ elif cat == "sneaker":
20
+ draw.polygon(P([(4, 34), (6, 22), (16, 20), (24, 28), (40, 31), (44, 36), (44, 41), (4, 41)]), fill=255)
21
+ elif cat == "bag":
22
+ draw.rectangle(P([(8, 18), (40, 42)]), fill=255)
23
+ draw.arc(P([(15, 6), (33, 30)]), 180, 360, fill=255, width=max(1, int(3 * k)))
24
+ elif cat == "hat":
25
+ draw.ellipse(P([(4, 30), (44, 40)]), fill=255)
26
+ draw.pieslice(P([(12, 10), (36, 46)]), 180, 360, fill=255)
27
+ elif cat == "chair":
28
+ draw.rectangle(P([(12, 6), (18, 42)]), fill=255)
29
+ draw.rectangle(P([(12, 24), (38, 30)]), fill=255)
30
+ draw.rectangle(P([(32, 28), (37, 42)]), fill=255)
31
+ elif cat == "lamp":
32
+ draw.polygon(P([(14, 8), (34, 8), (40, 22), (8, 22)]), fill=255)
33
+ draw.rectangle(P([(22, 22), (26, 38)]), fill=255)
34
+ draw.rectangle(P([(14, 38), (34, 42)]), fill=255)
35
+ elif cat == "bottle":
36
+ draw.rectangle(P([(20, 4), (28, 12)]), fill=255)
37
+ draw.polygon(P([(20, 12), (28, 12), (34, 20), (34, 43), (14, 43), (14, 20)]), fill=255)
38
+
39
+
40
+ def render(cat, color, pattern, rng):
41
+ """rng: numpy RandomState for jitter (position, scale, background, pattern phase, noise)."""
42
+ k = rng.uniform(0.82, 1.0)
43
+ ox, oy = rng.uniform(0, S - 48 * k + 1e-6), rng.uniform(0, S - 48 * k + 1e-6)
44
+ mask = Image.new("L", (S, S), 0)
45
+ _shape(ImageDraw.Draw(mask), cat, k, ox, oy)
46
+ m = np.asarray(mask, dtype=np.float64) / 255.0
47
+ bg = rng.uniform(214, 240)
48
+ img = np.ones((S, S, 3)) * np.array([bg, bg - rng.uniform(0, 8), bg - rng.uniform(0, 8)])
49
+ base = np.array(RGB[color], dtype=np.float64)
50
+ accent = np.where(base > 128, base * 0.45, np.minimum(255, base + 150))
51
+ yy, xx = np.mgrid[0:S, 0:S]
52
+ if pattern == "striped":
53
+ phase = rng.uniform(0, 6)
54
+ sel = ((xx + yy + phase) % 7) < 3
55
+ elif pattern == "dotted":
56
+ px, py = rng.uniform(0, 6), rng.uniform(0, 6)
57
+ sel = (((xx + px) % 7 - 3.5) ** 2 + ((yy + py) % 7 - 3.5) ** 2) < 3.2
58
+ else:
59
+ sel = np.zeros((S, S), bool)
60
+ fill = np.where(sel[:, :, None], accent, base)
61
+ img = img * (1 - m[:, :, None]) + fill * m[:, :, None]
62
+ # dark outline so white items stay visible
63
+ edge = (np.abs(np.diff(np.pad(m, 1), axis=0))[:-1, 1:-1] + np.abs(np.diff(np.pad(m, 1), axis=1))[1:-1, :-1]) > 0.3
64
+ img[edge] = img[edge] * 0.35
65
+ img += rng.normal(0, 3.0, img.shape)
66
+ return np.clip(np.round(img), 0, 255).astype(np.uint8)
selftest.json ADDED
The diff for this file is too large to render. See raw diff
 
text.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc1059e6665f011f5139d272a35384b9705073095e91c86ca5757239c3b17ffa
3
+ size 297112