flowseek / README.md
Flor's picture
card: update from fastforks 1cbf1671
614f7f5 verified
|
Raw History Blame Contribute Delete
8.04 kB
---
library_name: "fastforks"
pipeline_tag: "image-to-image"
license: "other"
license_name: "see-model-card"
license_link: "https://github.com/mattpoggi/flowseek/blob/ba6407370bceeb96244cad4d60d37489f6eb471e/scripts/get_weights.sh"
tags:
- "flow"
- "optical-flow"
- "motion-estimation"
- "cuda"
- "pytorch"
- "fastforks"
- "fastforks-card"
---
# FlowSeek (fastforks)
FlowSeek (Poggi & Tosi 2025), size T: SEA-RAFT (ResNet18 encoders, initial flow regression, ConvNeXt update block, 4 iterations) helped by a frozen Depth Anything V2 Small. Its features join the matching features, and its relative depth gives 8 motion bases (the flow fields that camera translations and rotations would cause), which a third encoder turns into extra context. Each image is resized to 518 x 518 for Depth Anything, whatever the input size. The authors' released weights; the network is ported from mattpoggi/flowseek. `iters=12` is the paper's FlowSeek (S). The larger M and L sizes use Depth Anything V2 Base, whose weights are non-commercial, and are not included.
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/mattpoggi/flowseek> at `ba6407370bce`.
Task: flow. Tier: experimental. Outputs: `flow`. Backends: `portable`.
Input: two RGB images (consecutive frames). Output: per-pixel motion from the first to the second (dx, dy in pixels).
## Variants
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/flow/models/flowseek/card.toml).
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|---|---|---|---|---|
| `flowseek-t@1` | [flowseek_T_TartanCT_TSKH.pth](https://drive.usercontent.google.com/download?id=1IQoyY5PpKSadtiGuhWwVCqvgD3y8CyFd&export=download&confirm=t) | `7913a91747a5` | Apache-2.0 AND BSD-3-Clause (assumed from code) | yes |
| `flowseek-t-things@1` | [flowseek_T_TartanCT.pth](https://drive.usercontent.google.com/download?id=1dnMzlRqX7wziynQvZjafsXjgbfibNAKG&export=download&confirm=t) | `d11f1969f933` | Apache-2.0 AND BSD-3-Clause (assumed from code) | yes |
| `flowseek-t-things-notartan@1` | [flowseek_T_CT.pth](https://drive.usercontent.google.com/download?id=1COOQFkMulzpBm4zMoWsaRGk7E3YcVr2I&export=download&confirm=t) | `d4bdd10dc5ad` | Apache-2.0 AND BSD-3-Clause (assumed from code) | yes |
## Licence
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0 AND BSD-3-Clause. Licence files: `LICENSE`, `LICENSE-BSD-3-Clause`.
Code licence: Apache-2.0 AND BSD-3-Clause; the authors do not state a separate weights licence.
Training data: TartanAir, FlyingChairs (Freiburg), Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), MPI Sintel optical flow, KITTI stereo / optical flow 2012 / 2015, HCI HD1K optical flow benchmark (TartanAir pretraining, FlyingChairs, FlyingThings3D (part of SceneFlow), then a mix of FlyingThings3D, Sintel, KITTI 2015 and HD1K (TSKH); the Depth Anything V2 Small inside was trained by its authors on synthetic and pseudo-labelled real images).
Licence evidence: <https://github.com/mattpoggi/flowseek/blob/ba6407370bceeb96244cad4d60d37489f6eb471e/scripts/get_weights.sh>
## Input size
any H x W >= 128, resized or padded inside the model
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): `flowseek-t` 1024x2048 (8.6 GiB peak allocated, 10.2 reserved); `flowseek-t-things` 1024x2048 (8.6 GiB peak allocated, 10.2 reserved); `flowseek-t-things-notartan` 1024x2048 (8.6 GiB peak allocated, 10.2 reserved)
## Speed and memory
### Flow, synthetic
| model | input | time (ms): original fp32 | original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `flowseek-t` | 1080x1920 | 650 | 457 | 395 | – | – | – | 8693 / 9880 | 8671 / 8712 | – |
| `flowseek-t-things` | 1080x1920 | 650 | 458 | 395 | – | – | – | 8693 / 9820 | 8671 / 8712 | – |
| `flowseek-t-things-notartan` | 1080x1920 | 651 | 458 | 396 | – | – | – | 8693 / 9820 | 8671 / 8712 | – |
### Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
| --- | --- | --- | --- | --- | --- | --- |
| `flowseek-t` | 0.87 | 0.18 (375x1242) | – | 0.63 | 0.46 (375x1242) | – |
| `flowseek-t-things` | 0.78 | 0.18 (375x1242) | – | 0.46 | 0.47 (375x1242) | – |
| `flowseek-t-things-notartan` | 0.81 | 0.18 (375x1242) | – | 0.46 | 0.46 (375x1242) | – |
### torch.compile and TensorRT
| 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) |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `flowseek-t` | 1080x1920 | 395 ms | 258 ms | 168 ms | – | 54.2 s | 107 s | 8671 / 8712 | 6126 / 6212 | – | 11328 / 11350 |
| `flowseek-t-things` | 1080x1920 | 395 ms | 258 ms | 168 ms | – | 54.6 s | 102 s | 8671 / 8712 | 6126 / 6212 | – | 11330 / 11352 |
| `flowseek-t-things-notartan` | 1080x1920 | 396 ms | 258 ms | 168 ms | – | 54.3 s | 97.5 s | 8671 / 8712 | 6126 / 6212 | – | 11328 / 11350 |
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
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.
## Accuracy
### On public datasets
Frames _10 -> _11 at full size against the flow_occ ground truth; EPE = mean of the per-pair mean end-point error, Fl-all = share of all valid pixels with an error > 3 px and > 5 % of the GT magnitude. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | note |
|---|---|---|---|---|---|
| `flowseek-t` | KITTI 2015 train, EPE / Fl-all % | 1.260 / 3.88 | 1.360 / 4.35 | 1.348 / 4.33 | trained on this dataset's public images; not a zero-shot result |
| `flowseek-t-things` | KITTI 2015 train, EPE / Fl-all % | 4.121 / 12.35 | 4.153 / 12.25 | 4.143 / 12.27 | |
| `flowseek-t-things-notartan` | KITTI 2015 train, EPE / Fl-all % | 3.923 / 12.83 | 3.904 / 12.97 | 3.933 / 12.97 | |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks c4a67979, 2026-11-11.
### Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.001 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.0077 and p99 <= 0.041 (absolute error of the output `flow`).
## Usage
```python
from fastforks import flow
model = flow.load("flowseek-t") # mode="auto": the preferred backend from recorded measurements
result = model("frame1.png", "frame2.png")
result.flow # [2, H, W] float32 tensor: dx, dy in input pixels
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
```
Install with `pip install fastforks` (see <https://github.com/kleinicke/fastforks>). Other variants: `flowseek-t`, `flowseek-t-things`, `flowseek-t-things-notartan`.
## Kernels backend
No kernels backend: no kernels mode was built for it; SEA-RAFT's engine and the Depth Anything kernels would fit and could be combined.
<!-- staged from fastforks 1cbf1671 (2026-11-15) -->