search

ShirAmir / dino-vit-features

Official implementation for the paper "Deep ViT Features as Dense Visual Descriptors".
https://dino-vit-features.github.io
MIT License
383 stars 44 forks source link

Supervised ViT Checkpoint #3

Closed mhamilton723 closed 2 years ago

mhamilton723 commented 2 years ago

Hi, thanks for this great repo! Could you by chance point me in the direction of the "Supervised ViT" described in Figure 3?

ShirAmir commented 2 years ago

Hi! As mentioned in the supplementary material, we used the supervised ViT vit_small_patch16_224 from the timm repository to generate Figure 3.

Our code supports running with different ViTs trained in a supervised manner / with DINO. Just specify --model_type <model_type> with any of the following options:

[dino_vits8 | dino_vits16 | dino_vitb8 | dino_vitb16 | vit_small_patch8_224 |  vit_small_patch16_224 | vit_base_patch8_224 | vit_base_patch16_224]

Also, we have just uploaded code to generate PCA to a collection of images as done in Figure 3. You can find it in pca.py.