Precomputed saliency maps for ImageNet-1k, generated using DeepGaze IIE (paper). These maps can be used for attention-guided sampling, foveated vision experiments, or any task requiring visual saliency information.
Saliency maps were extracted using DeepGaze IIE with the MIT1003 centerbias prior. Maps are saved at native resolution (matching the original ImageNet image dimensions) and normalized to probability… See the full description on the dataset page:
https://huggingface.co/datasets/Hafez/ImageNet1k-Saliency-Maps.
