timm

/

tf_efficientnetv2_l.in21k

like 0

Follow

PyTorch Image Models 736

Are there any examples available for implementing inference for this model, with label descriptions? I found ImageNetInfo in the source code and gave that a try:

        indices = top5_class_indices.cpu().numpy()

        subset = infer_imagenet_subset(model)
        info = ImageNetInfo(subset=subset)

        for index in indices[0]:
            print(info.index_to_description(index))

And this seems to be close to what I need, but the id2label set that it uses doesn't seem to line up with the actual indexes. The output I get for the example beignets image:

homo, man, human being, human
bird
world, human race, humanity, humankind, human beings, humans, mankind, man
equipment
animal, animate being, beast, brute, creature, fauna

I assume right now that I'm just using this wrong, so I'd love an example if one is available. I also tried building an id2label dict from huggingface/label-files

    repo_id = "huggingface/label-files"
    filename = "imagenet-22k-id2label.json"
    id2label = json.load(open(hf_hub_download(repo_id, filename, repo_type="dataset"), "r"))
    id2label = {int(k): v for k, v in id2label.items()}

But this yielded identical results