Ballon999 commited on
Commit
2eeebc7
·
verified ·
1 Parent(s): d8e9e29

docs: update cellpose-js links to TheJacksonLaboratory/cellpose-js

Browse files
Files changed (1) hide show
  1. README.md +7 -7
README.md CHANGED
@@ -24,7 +24,7 @@ language:
24
  Single-file FP16 ONNX export of **Cellpose-SAM** (CPSAM), the ViT-L–based cellular
25
  segmentation model from [Stringer et al., 2025](https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1).
26
  Intended for in-browser inference via WebGPU through
27
- [`cellpose-js`](https://github.com/belkassaby/Cellpose.js), but usable from any
28
  ONNX Runtime backend that supports opset 18 and FP16 graph IO.
29
 
30
  - **Architecture:** SAM ViT-L image encoder (modified — patch size 8, no windowed
@@ -84,7 +84,7 @@ const result = await cp.segment(
84
  The first call fetches 588 MB from the Hub; subsequent calls hit IndexedDB and
85
  cold-start in under ~2 s.
86
 
87
- See the [cellpose-js README](https://github.com/belkassaby/Cellpose.js#readme)
88
  for the full API, parameter reference, and Python parity notes.
89
 
90
  ### With ONNX Runtime directly (Python)
@@ -101,7 +101,7 @@ flow_y, flow_x, cellprob = out[0, 0], out[0, 1], out[0, 2]
101
 
102
  For flow-dynamics postprocessing (Euler integration → convergence clustering →
103
  connected components → size/flow filtering), use either:
104
- - the JS port in [`cellpose-js/src/dynamics`](https://github.com/belkassaby/Cellpose.js/tree/main/src), or
105
  - the original Python implementation in
106
  [`cellpose.dynamics`](https://cellpose.readthedocs.io/) — input/output
107
  contracts match.
@@ -139,8 +139,8 @@ ORT-web 1.26 is ~2.3× faster than 1.20 on the WebGPU kernels — the steady-sta
139
  ## How the model was generated
140
 
141
  The export path is documented in
142
- [`docs/STAGE0-RESULTS.md`](https://github.com/belkassaby/Cellpose.js/blob/main/docs/STAGE0-RESULTS.md)
143
- and [`docs/PLAN.md §1.5, §2`](https://github.com/belkassaby/Cellpose.js/blob/main/docs/PLAN.md).
144
  The short version:
145
 
146
  1. **Source weights**: `mouseland/cellpose-sam` (PyTorch, 1.23 GB, 304.6 M params).
@@ -243,7 +243,7 @@ https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1
243
 
244
  - **Source checkpoint:** `mouseland/cellpose-sam` on Hugging Face Hub.
245
  - **Export scripts and ONNX artifact ETag** (`52fd6881…`) are recorded in
246
- [`docs/STAGE0-RESULTS.md`](https://github.com/belkassaby/Cellpose.js/blob/main/docs/STAGE0-RESULTS.md).
247
  - **Parity test fixtures** (numpy-generated FP32 reference tiles + expected
248
  flow outputs) live in `tests/fixtures/` in the `cellpose-js` repo.
249
 
@@ -263,7 +263,7 @@ https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1
263
 
264
  ## Related
265
 
266
- - **Code:** [`belkassaby/Cellpose.js`](https://github.com/belkassaby/Cellpose.js) — TypeScript inference + dynamics port.
267
  - **npm:** [`cellpose-js`](https://www.npmjs.com/package/cellpose-js).
268
  - **Upstream:** [`MouseLand/cellpose`](https://github.com/MouseLand/cellpose) and [`mouseland/cellpose-sam`](https://huggingface.co/mouseland/cellpose-sam).
269
  - **Paper:** [Cellpose-SAM (bioRxiv 2025.04.28)](https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1).
 
24
  Single-file FP16 ONNX export of **Cellpose-SAM** (CPSAM), the ViT-L–based cellular
25
  segmentation model from [Stringer et al., 2025](https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1).
26
  Intended for in-browser inference via WebGPU through
27
+ [`cellpose-js`](https://github.com/TheJacksonLaboratory/cellpose-js), but usable from any
28
  ONNX Runtime backend that supports opset 18 and FP16 graph IO.
29
 
30
  - **Architecture:** SAM ViT-L image encoder (modified — patch size 8, no windowed
 
84
  The first call fetches 588 MB from the Hub; subsequent calls hit IndexedDB and
85
  cold-start in under ~2 s.
86
 
87
+ See the [cellpose-js README](https://github.com/TheJacksonLaboratory/cellpose-js#readme)
88
  for the full API, parameter reference, and Python parity notes.
89
 
90
  ### With ONNX Runtime directly (Python)
 
101
 
102
  For flow-dynamics postprocessing (Euler integration → convergence clustering →
103
  connected components → size/flow filtering), use either:
104
+ - the JS port in [`cellpose-js/src/dynamics`](https://github.com/TheJacksonLaboratory/cellpose-js/tree/main/src), or
105
  - the original Python implementation in
106
  [`cellpose.dynamics`](https://cellpose.readthedocs.io/) — input/output
107
  contracts match.
 
139
  ## How the model was generated
140
 
141
  The export path is documented in
142
+ [`docs/STAGE0-RESULTS.md`](https://github.com/TheJacksonLaboratory/cellpose-js/blob/main/docs/STAGE0-RESULTS.md)
143
+ and [`docs/PLAN.md §1.5, §2`](https://github.com/TheJacksonLaboratory/cellpose-js/blob/main/docs/PLAN.md).
144
  The short version:
145
 
146
  1. **Source weights**: `mouseland/cellpose-sam` (PyTorch, 1.23 GB, 304.6 M params).
 
243
 
244
  - **Source checkpoint:** `mouseland/cellpose-sam` on Hugging Face Hub.
245
  - **Export scripts and ONNX artifact ETag** (`52fd6881…`) are recorded in
246
+ [`docs/STAGE0-RESULTS.md`](https://github.com/TheJacksonLaboratory/cellpose-js/blob/main/docs/STAGE0-RESULTS.md).
247
  - **Parity test fixtures** (numpy-generated FP32 reference tiles + expected
248
  flow outputs) live in `tests/fixtures/` in the `cellpose-js` repo.
249
 
 
263
 
264
  ## Related
265
 
266
+ - **Code:** [`TheJacksonLaboratory/cellpose-js`](https://github.com/TheJacksonLaboratory/cellpose-js) — TypeScript inference + dynamics port.
267
  - **npm:** [`cellpose-js`](https://www.npmjs.com/package/cellpose-js).
268
  - **Upstream:** [`MouseLand/cellpose`](https://github.com/MouseLand/cellpose) and [`mouseland/cellpose-sam`](https://huggingface.co/mouseland/cellpose-sam).
269
  - **Paper:** [Cellpose-SAM (bioRxiv 2025.04.28)](https://www.biorxiv.org/content/10.1101/2025.04.28.651001v1).