isarandi / tensorflow-inputs Goto Github PK

Example

frame_paths = glob(f'/globalwork/datasets/3DPW/imageFiles/courtyard_arguing_00/image_*.jpg')
batches_gpu, batches_cpu = tfinp.image_files(frame_paths, batch_size=64, prefetch_gpu=2, tee_cpu=True)
batch_gpu, batch_cpu = next(iter(zip(batches_gpu, batches_cpu)))

Expected behavior

One batch is fetched

Actual behavior

An error occurs

TypeError: generator yielded an element of shape (1920, 1080, 3) where an element of shape (1080, 1920, 3) was expected

Reason

The image_files function passes [width, height] on as imshape, which is used for output_shapes with Dataset.from_generator while cv2.imread will return images as [height, width, channels]. MeTRAbs seems to expect the input format as [height, width] thus imshape in image_files should probably change.