rdancer-ComfyUI_Florence2SAM2

T

Jan Minar 487b0510ab fixes (hopefully): torch.OutOfMemoryError: Allocation on device

- Fix the SAM2 model being loaded on every run due to string being compared to
  class instance
- Manually unload the old SAM2 model from VRAM before loading the new one
- For a good measure, garbage-collect the RAM version as well

2024-10-09 20:17:31 +01:00

configs

SAM2 added

2024-08-01 12:47:29 +02:00

models/sam2

Fix model paths

2024-09-30 03:04:22 +00:00

test

Fix model paths

2024-09-30 03:04:22 +00:00

utils

Update sam.py -- third time's the charm

2024-10-02 22:51:17 +01:00

videos

example videos

2024-08-06 00:39:38 +02:00

workflows

allow to select device

2024-10-04 01:27:36 +01:00

__init__.py

allow to select device

2024-10-04 01:27:36 +01:00

.gitattributes

add mp4 to lfs

2024-08-06 00:39:11 +02:00

.gitignore

working from the commandline

2024-09-29 20:39:16 +00:00

app.py

fixes (hopefully): torch.OutOfMemoryError: Allocation on device

2024-10-09 20:17:31 +01:00

README.md

Update README.md

2024-09-30 04:13:50 +01:00

requirements.txt

Build dependency; getting errors so try putting it in requirements.txt

2024-10-03 20:29:27 +01:00

test.py

working from the commandline

2024-09-29 20:39:16 +00:00

README.md

ComfyUI custom node implementing Florence 2 + Segment Anything Model 2, based on SkalskiP's HuggingFace space

RdancerFlorence2SAM2GenerateMask - the node is self-contained, and does not require separate model loaders. Models are lazy-loaded, and cached. Model unloading, if required, must be done manually.

Testing

Run python test.py test/*_source.png "products" 👉 the resulting images must pixel-match test/*output*.