vit_smash

helper code

    model = CustomViTRegressor(should_load_from_disk=False)
    model.update_model_from_checkpoint(2)
    custom_data_collator = custom_data_collator_function(ViTImageProcessor())
    train_loader = DataLoader(dataset["train"], batch_size=32, collate_fn=custom_data_collator)
    model = model.to(device)

and here are some helpful functions

def custom_data_collator_function(processor):
    def return_func(batch):
        images = [item["image"] for item in batch]

        inputs = processor(images, return_tensors="pt")
        targets = torch.tensor([[
            item["Start"],
            item["A"],
            item["B"],
            item["X"],
            item["Y"],
            item["Z"],
            item["DPadUp"],
            item["DPadDown"],
            item["DPadLeft"],
            item["DPadRight"],
            item["L"],
            item["R"],
            item["LPressure"] / 255,
            item["RPressure"] / 255,
            item["XAxis"] / 255,
            item["YAxis"] / 255,
            item["CXAxis"] / 255,
            item["CYAxis"] / 255] for item in batch], dtype=torch.float32)

        return inputs, targets

    return return_func


def is_interesting_target(target):
    if target[0] == 1 or \
            target[1] == 1 or \
            target[2] == 1 or \
            target[3] == 1 or \
            target[4] == 1 or \
            target[5] == 1 or \
            target[6] == 1 or \
            target[7] == 1 or \
            target[8] == 1 or \
            target[9] == 1 or \
            target[10] == 1 or \
            target[11] == 1:
        return 1
    if target[12] >= THRESHOLD or \
            target[13] >= THRESHOLD or \
            abs(target[14] - 127.5) >= THRESHOLD or \
            abs(target[15] - 127.5) >= THRESHOLD or \
            abs(target[16] - 127.5) >= THRESHOLD or \
            abs(target[17] - 127.5) >= THRESHOLD:
        return 1
    return 0