Output of model

[AnhMinhTran]

Is the output of timm models a list? I have been trying to use timm models as the backbone for Torch Vision Faster RCNN, they required their backbone output to be an ordered dict, is there any conversion/parameters that I am missing in here? Please see my code below

    model_backbone_name = "tf_efficientnetv2_b3"
    checkpoint_path = "./models/model_best.pth.tar"
    checkpoint_path_url = pathlib.Path(checkpoint_path).resolve().as_uri()
    efficientnet.default_cfgs[model_backbone_name]["url"] = checkpoint_path_url
    backbone = timm.create_model(
        model_backbone_name,
        features_only=True,
        num_classes=0,
        global_pool='')
    backbone.out_channels = backbone.feature_info[-1]["num_chs"]
    anchor_generator = AnchorGenerator(
        sizes=((16, 32, 64, 128, 256),), aspect_ratios=((0.25, 0.5, 1.0, 2.0),)
    )
    roi_pooler = torchvision.ops.MultiScaleRoIAlign(
        featmap_names=["0"], output_size=7, sampling_ratio=2
    )
    fastercnn_model = FasterRCNN(
        backbone=backbone,
        num_classes=9,
        rpn_anchor_generator=anchor_generator,
        box_roi_pool=roi_pooler,
        min_size=512,
        max_size=512
    )
    for idx, child in enumerate(fastercnn_model.children()):
        if idx == 1:
            # Now set requires_grad for that idx to False
            for param in child.parameters():
                param.requires_grad = False
            break
    return fastercnn_model

Error

[yohann84L]

You set features_only=True in timm.create_model(), thus the model will return a list of features. See Turn any model into a feature extractor from fastai timm doc for more explanation.

So you need to create the backbone like this:

backbone = timm.create_model(
    model_backbone_name,
    num_classes=0,  # this is important to remove fc layers
    global_pool=""  # this is important to remove fc layers
)