The evaluation code is in cluster_prob.py, linear_prob_cocostuff.py and linear_prob_voc.py. The entry point of self supervised training with flow is dino_training/train.py. The overall code is refactored a bit but shall be all right, but I haven't got the time to fully organize the training data, will do it later.
To run evaluation, download pretrained weights, then
# Linear Probing on COCOStuff and VOC
python3 ./linear_prob_cocostuff.py --network "network.dino:vit_base_8" \
--checkpoint ./weights/b8_flow_dino.pth
python3 ./linear_prob_voc.py --network "network.dino:vit_small_8" \
--checkpoint ./weights/s8_flow_dino.pth
# Cluster Probing on VOC
python3 ./cluster_prob.py --network "network.dino:vit_small_16" \
--checkpoint ./weights/s16_flow_dino.pth