feat(cellpose): update to v4.0.8 with GPU support

[an-altosian]

Summary

• Upgrade cellpose from 3.1.0 to 4.0.8 (includes Cellpose-SAM integration)
• Replace biocontainers/cellpose:3.1.0_cv1 with GPU-enabled dongzehe/cellpose-gpu:4.0.8
• Add GPU support via process_gpu label and task.ext.use_gpu flag
• Change model input from path to val (pretrained model name string)
• Add cells output (*seg.npy) for downstream analysis
• Outputs moved to ${prefix}/ directory for consistency
• Add Dockerfile for reproducible container builds

Changes

File	Change
main.nf	Upgrade to v4.0.8, GPU container, val model input, task.ext.use_gpu, cells output
meta.yml	Match new interface: model as val string, add cells output
Dockerfile	New: GPU container based on pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime
tests/main.nf.test	Update inputs for new interface
tests/main.nf.test.snap	Updated snapshots for v4.0.8 outputs

Container

dongzehe/cellpose-gpu:4.0.8 (hosted on quay.io, public)

• Base: pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime
• Packages: cellpose==4.0.8, numpy, scipy, matplotlib, scikit-image, opencv-python-headless
• Falls back to CPU automatically when no GPU is available

GPU control

withName: 'CELLPOSE' {
    ext.use_gpu = true  // passes --use_gpu to cellpose CLI
}

When task.ext.use_gpu is not set or false, cellpose runs on CPU only.

Usage

CELLPOSE(
    ch_images,      // tuple val(meta), path(image)
    "cyto3"         // model name: 'cyto3', 'nuclei', 'CPSAM', or ''
)

Licensing

• Cellpose: BSD-3-Clause
• SAM (segment_anything): Apache 2.0
• Model weights NOT bundled (downloaded at runtime)

🤖 Generated with Claude Code

[an-altosian]

Hi there, I saw that the cellpose module was updated three weeks ago but is still on cellpose 3. Here is a draft of updating it to cellpose v4 with GPU support, which has its cellpose-SAM based model compatibility. It works great on my side.