VisionHOPE pretrained weights

Pretrained weights for the hierarchical VisionHOPE-T/S/B models.

• ImageNet-1K: classification checkpoints for Tiny, Small, and Base.
• COCO: Mask R-CNN checkpoints under the 1× schedule for Tiny, Small, and Base, and the 3× schedule for Tiny and Small.
• ADE20K: UPerNet checkpoints for Tiny, Small, and Base.

Download the .pth files from Assets below.

Paper · Hugging Face mirror