chou141253 / fgvc-herbs Goto Github PK

I noticed that your article mentions about resnet, can you provide the content about resnet?

Is this project only support swin-transformer?
I wanna to use this model to train CNN, what should I do?

There is no eval.py file, and there is no 'suppression' function in you eval.py file from PIM project

How do you obtain the final inference result?I find your best_ Top1 is the highest accuracy value from all classifiers.

Hi,
I tried to reproduce your results and encountered an error when running run_evaluation.py.
The backbone and return_nodes parameters are missing.

If someone encountered the same problem too, change the "build_model" function to be:

def build_model(pretrainewd_path: str,
                img_size: int,
                fpn_size: int,
                num_classes: int,
                num_selects: dict,
                use_fpn: bool = True,
                use_selection: bool = True,
                use_combiner: bool = True,
                comb_proj_size: int = None):
    from models.pim_module.pim_module import PluginMoodel

    # swin_base_patch4_window12_384_in22k
    backbone = create_model('swin_large_patch4_window12_384_in22k', pretrained=True)

    model = \
        PluginMoodel(
            backbone=backbone,
            return_nodes=None,
            img_size=img_size,
            use_fpn=use_fpn,
            fpn_size=fpn_size,
            proj_type="Linear",
            upsample_type="Conv",
            use_selection=use_selection,
            num_classes=num_classes,
            num_selects=num_selects,
            use_combiner=use_combiner,
            comb_proj_size=comb_proj_size)

    if pretrainewd_path != "":
        print("loaded model")
        ckpt = torch.load(pretrainewd_path)
        model.load_state_dict(ckpt['model'])

    model.eval()

    return model

Hi, why does it take me hours to run one epoch.

Hello, author. When I was using your code to generate a heat map and test my own trained model, the console displayed the following error. How can I solve it?

init() missing 2 required positional arguments: 'backbone' and 'return_nodes'

I would greatly appreciate it if you could help me solve this problem!

Congratulations, your algorithm has achieved the best accuracy on two bird data sets. Which backbone network is it based on? Swin-T or Swin-L ? Looking forward to your reply!

Hi, thank you very much for the code. When I want to run python main.py with CUB-200-2011 dataset --c . /configs/config.yaml, I realized that the CUB-200-2011 dataset does not have a
tain/
│ ├── class1/
│ │ ├── img001.jpg
│ │ ├── img002.jpg
│ │ └── ....
│ ├── class2/ │ ├── class2/
│ │ ├── img001.jpg
│ │ ├── img002.jpg
I would like to ask if these training images are processed by myself, or where can I download them directly?

Hello, author. Which one should be selected as the output?