I’m looking for PyTorch weights for any semi-modern CNN architectures (ResNet’s, etc.) pretrained on Open Images V4 or V5. If anyone has seen this or has trained on it themselves and is willing to share, this would be greatly appreciated!
Side note: I noticed that the TensorFlow object detection API has detection models pretrained on OIV4. Is it possible/practical to pull the weights for just the CNN backbone of one of these and then convert this to a format that PyTorch can load?