Video Generation Models are General-Purpose Vision Learners

Benchmark Model Rank Results
3d-human-pose-estimation-on-emdbGenCeption Specialist-L–Average MPJPE (mm): 71.8
camera-pose-estimation-on-sintelGenCeption Specialist-L–ATE: 0.050RPE-T: 0.018RPE-R: 0.464
referring-video-object-segmentation-on-refvos-davisGenCeption Specialist-L–J&F: 76.4
surface-normals-estimation-on-sintelGenCeption Generalist-L–mAE: 29.3