card: update from fastforks 1cbf1671
Browse files
README.md
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: "fastforks"
|
| 3 |
+
pipeline_tag: "keypoint-detection"
|
| 4 |
+
license: "apache-2.0"
|
| 5 |
+
tags:
|
| 6 |
+
- "tracking"
|
| 7 |
+
- "point-tracking"
|
| 8 |
+
- "tracking-any-point"
|
| 9 |
+
- "video"
|
| 10 |
+
- "cuda"
|
| 11 |
+
- "pytorch"
|
| 12 |
+
- "fastforks"
|
| 13 |
+
- "fastforks-card"
|
| 14 |
+
---
|
| 15 |
+
# TAPNext (fastforks)
|
| 16 |
+
|
| 17 |
+
TAPNext (Zholus, Doersch, Yang, Koppula et al., Google DeepMind 2025): tracking any point as next-token prediction. A TRecViT-B (12 blocks, each a gated linear recurrence over time per token and a ViT layer over the tokens of a frame) reads 8 x 8 patches of 256 x 256 frames plus one token per query point and predicts, frame by frame, the point's position (512-bin classification, windowed soft-argmax) and whether it is visible. Online: each frame only sees the past.
|
| 18 |
+
|
| 19 |
+
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: <https://github.com/kleinicke/fastforks>. Kernels for the `kernels` backend: <https://huggingface.co/fastforks/kernels>. Original code: <https://github.com/google-deepmind/tapnet> at `ceedd60bb9f6`.
|
| 20 |
+
|
| 21 |
+
Task: tracking. Tier: experimental. Outputs: `tracks`, `visibility`. Backends: `portable`, `kernels`.
|
| 22 |
+
|
| 23 |
+
Input: a video (its frames in order) and query points `(t, x, y)`. Output: the position of every query in every frame and the probability that it is visible there.
|
| 24 |
+
|
| 25 |
+
## Variants
|
| 26 |
+
|
| 27 |
+
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in [`card.toml`](https://github.com/kleinicke/fastforks/blob/main/src/fastforks/tracking/models/tapnext/card.toml).
|
| 28 |
+
|
| 29 |
+
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|
| 30 |
+
|---|---|---|---|---|
|
| 31 |
+
| `tapnext@1` | [bootstapnext_ckpt.npz](https://storage.googleapis.com/dm-tapnet/tapnext/bootstapnext_ckpt.npz) | `19de6b149555` | Apache-2.0 | yes |
|
| 32 |
+
| `tapnext-pp@1` | [tapnextpp_ckpt.pt](https://storage.googleapis.com/dm-tapnet/tapnextpp/tapnextpp_ckpt.pt) | `cb96a43444cc` | Apache-2.0 | yes |
|
| 33 |
+
|
| 34 |
+
## Licence
|
| 35 |
+
|
| 36 |
+
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Licence files: `LICENSE`.
|
| 37 |
+
|
| 38 |
+
Training data: Upstream: TAPNext is trained on synthetic Kubric MOVi-F sequences and fine-tuned with BootsTAP self-training on unlabelled real videos (not released); TAPNext++ is further fine-tuned on PointOdyssey and Kubric-1024 (1024-frame synthetic sequences). None has a dataset entry here yet..
|
| 39 |
+
|
| 40 |
+
Licence evidence: <https://github.com/google-deepmind/tapnet/blob/ceedd60bb9f6c9f7c74796e4445aff6c6ddf7c9b/README.md#license-and-disclaimer>
|
| 41 |
+
|
| 42 |
+
## Input size
|
| 43 |
+
|
| 44 |
+
any H x W >= 32, padded to a multiple of 32
|
| 45 |
+
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `tapnext` 3000x4000 (4.3 GiB peak allocated, 4.5 reserved); `tapnext-pp` 3000x4000 (4.3 GiB peak allocated, 4.5 reserved)
|
| 46 |
+
|
| 47 |
+
## Speed and memory
|
| 48 |
+
|
| 49 |
+
### Tracking, synthetic, 480x854
|
| 50 |
+
|
| 51 |
+
| model | time (ms): original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|
| 52 |
+
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 53 |
+
| `tapnext` | 909 | 529 | 382 | 2.4x | 1.4x | 1721 / 1902 | 1220 / 1294 | 1313 / 1374 |
|
| 54 |
+
| `tapnext-pp` | 1493 | 529 | β | β | β | 1527 / 1718 | 1220 / 1294 | β |
|
| 55 |
+
|
| 56 |
+
### Startup and first call
|
| 57 |
+
|
| 58 |
+
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|
| 59 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 60 |
+
| `tapnext` | 16.92 | 2.72 | 2.63 | 2.17 | 0.69 | 0.56 |
|
| 61 |
+
| `tapnext-pp` | 4.83 | 1.34 | β | 1.80 | 0.77 | β |
|
| 62 |
+
|
| 63 |
+
### torch.compile and TensorRT
|
| 64 |
+
|
| 65 |
+
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
|
| 66 |
+
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|
| 67 |
+
| `tapnext` | 480x854 | 529 ms | 398 ms | deferred (build >5 min) | 382 ms | 40.5 s | β | 1220 / 1294 | 1164 / 1272 | 1313 / 1374 | β |
|
| 68 |
+
| `tapnext-pp` | 480x854 | 529 ms | planned for 0.2.1 | β | β | β | β | 1220 / 1294 | β | β | β |
|
| 69 |
+
|
| 70 |
+
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
|
| 71 |
+
|
| 72 |
+
Hardware, protocol and caveats: [benchmarks](https://github.com/kleinicke/fastforks/blob/main/docs/benchmarks.md) (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or `(fp32)` where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
|
| 73 |
+
|
| 74 |
+
## Accuracy
|
| 75 |
+
|
| 76 |
+
### On public datasets
|
| 77 |
+
|
| 78 |
+
TAP-Vid protocol: frames resized to 256 x 256 (PIL Lanczos) before the model (AllTracker: the original frames, resized by the model), one query per track at its first visible point, tracked online (TAPNext, Track-On) or over the whole clip (TAPIR, AllTracker, LocoTrack); AJ, delta_avg (< 1, 2, 4, 8, 16 px at 256 x 256) and occlusion accuracy averaged over the videos. Original = the original code named under Details, same checkpoint and protocol.
|
| 79 |
+
|
| 80 |
+
| variant | dataset | original fp32 | original fp16 | portable | kernels | note |
|
| 81 |
+
|---|---|---|---|---|---|---|
|
| 82 |
+
| `tapnext` | TAP-Vid DAVIS (query first), every 3rd video, AJ / delta_avg / OA % | 71.01 / 82.01 / 95.18 | 71.02 / 82.02 / 95.19 | 71.02 / 82.02 / 95.19 | 71.02 / 82.03 / 95.23 | |
|
| 83 |
+
| `tapnext` | TAP-Vid DAVIS (query first), AJ / delta_avg / OA % | 66.55 / 79.48 / 92.16 | 66.55 / 79.49 / 92.17 | 66.55 / 79.49 / 92.17 | 66.56 / 79.50 / 92.18 | |
|
| 84 |
+
| `tapnext-pp` | TAP-Vid DAVIS (query first), every 3rd video, AJ / delta_avg / OA % | 71.48 / 82.43 / 94.96 | 71.50 / 82.43 / 94.96 | 71.50 / 82.43 / 94.96 | 71.47 / 82.41 / 94.98 | |
|
| 85 |
+
|
| 86 |
+
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-12.
|
| 87 |
+
|
| 88 |
+
### Parity with the original code
|
| 89 |
+
|
| 90 |
+
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 4.6e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0063 and p99 <= 0.036 (absolute error of the output `tracks`).
|
| 91 |
+
|
| 92 |
+
## Limitations
|
| 93 |
+
|
| 94 |
+
- Online, causal: a point is tracked from its query frame on; in frames before it the model has no information (the tracks there are its guess, with a low visibility). Query on the first frame, or run the reversed video for the backward part.
|
| 95 |
+
- Frames are resized to 256 x 256 (the training size), whatever their size and aspect ratio.
|
| 96 |
+
- One video per call; all queries of a call share the video.
|
| 97 |
+
|
| 98 |
+
## Usage
|
| 99 |
+
|
| 100 |
+
```python
|
| 101 |
+
from fastforks import tracking
|
| 102 |
+
|
| 103 |
+
model = tracking.load("tapnext") # mode="auto": the preferred backend from recorded measurements
|
| 104 |
+
result = model(frames, [[0, x, y]]) # frames: the video's images in order; queries (t, x, y)
|
| 105 |
+
result.tracks # [T, Q, 2] float32: x, y in input pixels; result.visible [T, Q] bool
|
| 106 |
+
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
|
| 107 |
+
```
|
| 108 |
+
|
| 109 |
+
Install with `pip install fastforks` (see <https://github.com/kleinicke/fastforks>). Other variants: `tapnext`, `tapnext-pp`.
|
| 110 |
+
|
| 111 |
+
## Kernels backend
|
| 112 |
+
|
| 113 |
+
`tracking.load("tapnext", mode="kernels")` runs our CUDA kernels (`fvm_conv`). It needs a prebuilt kernel for your GPU architecture (downloaded from <https://huggingface.co/fastforks/kernels> on first use) and a recent NVIDIA driver; if there is none it raises `BackendUnavailableError` with the reason and never falls back silently. `mode="auto"` picks `kernels` when it can run and logs the choice.
|
| 114 |
+
|
| 115 |
+
<!-- staged from fastforks 1cbf1671 (2026-11-15) -->
|