search

fanq15 / FSOD-code

MIT License
243 stars 61 forks source link

mask rcnn? #9

Open cjpark87 opened 4 years ago

cjpark87 commented 4 years ago

Thanks for open sourcing your repo. I would be interested in using FSOD with mask rcnn. I think the instruction in the README is about object detection, rather than image segmentation. Could you provide instruction on building it with mask rcnn?

cjpark87 commented 4 years ago

Maybe I can follow https://github.com/roytseng-tw/Detectron.pytorch/blob/master/configs/baselines/e2e_mask_rcnn_R-50-C4_2x.yaml?

It would be great if you can upload your configuration yaml file for mask rcnn!

TachibanaYoshino commented 4 years ago

In the train_net_step.py, in the 285 line , the code show as below：

Model

```

maskRCNN = Generalized_RCNN()

if cfg.CUDA:
    maskRCNN.cuda()

### Optimizer ###
gn_param_nameset = set()
for name, module in maskRCNN.named_modules():
    if isinstance(module, nn.GroupNorm):
        gn_param_nameset.add(name+'.weight')
        gn_param_nameset.add(name+'.bias')
   ...........


**Is fsod a dataset for few shot object detection, and also a dataset for few shot image segmentation?** Does it have segmentation labels?  In the train_net_step.py,  the model is mask rcnn, not faster rcnn ???
fanq15 commented 4 years ago

Actually, instance segmentation is out of our scope. I removed those useless configs from the repo, you can use these configs in detectron-tw to construct your mask-rcnn version.