Linwei-Chen commited on
Commit
a56c8ba
·
verified ·
1 Parent(s): d4c6267

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +17 -7
README.md CHANGED
@@ -1,3 +1,7 @@
 
 
 
 
1
  # SFM: Spatial Frequency Modulation for Semantic Segmentation (TPAMI 2025)
2
 
3
  Official checkpoints for the paper **Spatial Frequency Modulation for Semantic Segmentation**
@@ -15,9 +19,10 @@ Two released technical lines:
15
  weighted alignment, on Mask2Former (mmsegmentation 1.x).
16
  Code and config: [`SFM_mask2former/`](https://github.com/Linwei-Chen/SFM/tree/main/SFM_mask2former).
17
 
18
- Every checkpoint below is **verified from training/eval logs to meet or exceed the
19
- paper-reported mIoU**. Each model folder contains `config.py` (the exact config paired
20
- with that checkpoint) and the weight file.
 
21
 
22
  ## Line 1 — Cityscapes (val mIoU, ResNet-50)
23
 
@@ -63,14 +68,19 @@ Weight file: `best_mIoU_iter_150000.pth`.
63
  for environment setup and qhull compilation):
64
 
65
  ```bash
 
66
  PYTHONPATH=<SFM_mmseg0x> python tools/test.py \
67
  <model_folder>/config.py <model_folder>/iter_40000.pth --eval mIoU
68
  ```
69
 
70
- **Line 2** (see [`SFM_mask2former/`](https://github.com/Linwei-Chen/SFM/tree/main/SFM_mask2former)):
71
- download the checkpoint, point `load_from` in
72
- [`mask2former_swin-t_8xb2-160k_ade20k-512x512_SFMv2.py`](https://github.com/Linwei-Chen/SFM/blob/main/SFM_mask2former/configs/mask2former/mask2former_swin-t_8xb2-160k_ade20k-512x512_SFMv2.py)
73
- to it, then run the standard mmseg 1.x `tools/test.py` with the config.
 
 
 
 
74
 
75
  ## Citation
76
 
 
1
+ ---
2
+ license: apache-2.0
3
+ ---
4
+
5
  # SFM: Spatial Frequency Modulation for Semantic Segmentation (TPAMI 2025)
6
 
7
  Official checkpoints for the paper **Spatial Frequency Modulation for Semantic Segmentation**
 
19
  weighted alignment, on Mask2Former (mmsegmentation 1.x).
20
  Code and config: [`SFM_mask2former/`](https://github.com/Linwei-Chen/SFM/tree/main/SFM_mask2former).
21
 
22
+ Every checkpoint below is **verified from training/eval logs**: each measured mIoU either
23
+ exceeds the paper-reported number or matches it after one-decimal rounding
24
+ (e.g. PSPNet-R50 D16 measures 78.05, rounding to the paper's 78.1). Each model folder
25
+ contains `config.py` (the exact config paired with that checkpoint) and the weight file.
26
 
27
  ## Line 1 — Cityscapes (val mIoU, ResNet-50)
28
 
 
68
  for environment setup and qhull compilation):
69
 
70
  ```bash
71
+ cd SFM_mmseg0x # from the GitHub repo checkout
72
  PYTHONPATH=<SFM_mmseg0x> python tools/test.py \
73
  <model_folder>/config.py <model_folder>/iter_40000.pth --eval mIoU
74
  ```
75
 
76
+ **Line 2** (code: [`SFM_mask2former/`](https://github.com/Linwei-Chen/SFM/tree/main/SFM_mask2former),
77
+ mmsegmentation 1.x):
78
+
79
+ ```bash
80
+ cd SFM_mask2former
81
+ python tools/test.py \
82
+ <model_folder>/config.py <model_folder>/best_mIoU_iter_150000.pth
83
+ ```
84
 
85
  ## Citation
86