{ "data": [ { "caption": "00 Val-00: W", "image_path": "path_to_validation/000002.jpg", "action_path": "path_to_validation/W.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "01 Val-01: S", "image_path": "path_to_validation/000003.jpg", "action_path": "path_to_validation/S.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "02 Val-02: A", "image_path": "path_to_validation/000004.jpg", "action_path": "path_to_validation/A.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "03 Val-03: D", "image_path": "path_to_validation/000005.jpg", "action_path": "path_to_validation/D.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "04 Val-04: u", "image_path": "path_to_validation/000000.jpg", "action_path": "path_to_validation/u.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "05 Val-05: d", "image_path": "path_to_validation/000001.jpg", "action_path": "path_to_validation/d.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "06 Val-06: l", "image_path": "path_to_validation/000006.jpg", "action_path": "path_to_validation/l.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "07 Val-07: r", "image_path": "path_to_validation/000007.jpg", "action_path": "path_to_validation/r.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "08 Val-00: key rand", "image_path": "path_to_validation/000002.jpg", "action_path": "path_to_validation/key_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "09 Val-01: key rand", "image_path": "path_to_validation/000003.jpg", "action_path": "path_to_validation/key_1_action_rand_2.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "10 Val-02: camera rand", "image_path": "path_to_validation/000004.jpg", "action_path": "path_to_validation/camera_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "11 Val-03: camera rand", "image_path": "path_to_validation/000005.jpg", "action_path": "path_to_validation/camera_1_action_rand_2.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "12 Val-00: key+camera excl rand", "image_path": "path_to_validation/000002.jpg", "action_path": "path_to_validation/key_camera_excl_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "13 Val-01: key+camera excl rand", "image_path": "path_to_validation/000003.jpg", "action_path": "path_to_validation/key_camera_excl_1_action_rand_2.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "14 Val-02: key+camera rand", "image_path": "path_to_validation/000004.jpg", "action_path": "path_to_validation/key_camera_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "15 Val-03: key+camera rand", "image_path": "path_to_validation/000005.jpg", "action_path": "path_to_validation/key_camera_1_action_rand_2.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "16 Val-04: (simultaneous) key rand", "image_path": "path_to_validation/000000.jpg", "action_path": "path_to_validation/key_2_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "17 Val-05: (simultaneous) camera rand", "image_path": "path_to_validation/000001.jpg", "action_path": "path_to_validation/camera_2_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "18 Val-06: (simultaneous) key+camera excl rand", "image_path": "path_to_validation/000006.jpg", "action_path": "path_to_validation/key_camera_excl_2_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "19 Val-07: (simultaneous) key+camera rand", "image_path": "path_to_validation/000007.jpg", "action_path": "path_to_validation/key_camera_2_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "20 Val-08: W+A", "image_path": "path_to_validation/000005.jpg", "action_path": "path_to_validation/WA.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "21 Val-09: S+u", "image_path": "path_to_validation/000013.jpg", "action_path": "path_to_validation/S_u.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "22 Val-08: Still", "image_path": "path_to_validation/000005.jpg", "action_path": "path_to_validation/still.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "23 Val-09: Still", "image_path": "path_to_validation/000013.jpg", "action_path": "path_to_validation/still.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "24 Val-06: key+camera excl rand Frame 4", "image_path": "path_to_validation/000006.jpg", "action_path": "path_to_validation/key_camera_excl_1_action_rand_1_f4.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "25 Val-07: key+camera excl rand Frame 4", "image_path": "path_to_validation/000007.jpg", "action_path": "path_to_validation/key_camera_excl_1_action_rand_2_f4.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "26 Train-00", "image_path": "path_to_validation/000500.jpg", "action_path": "path_to_validation/000500_action.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "27 Train-01", "image_path": "path_to_validation/001000.jpg", "action_path": "path_to_validation/001000_action.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "28 Doom-00: W", "image_path": "path_to_validation/000000.jpg", "action_path": "path_to_validation/W.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "29 Doom-01: key rand", "image_path": "path_to_validation/000001.jpg", "action_path": "path_to_validation/key_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "30 Doom-02: camera rand", "image_path": "path_to_validation/000002.jpg", "action_path": "path_to_validation/camera_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 }, { "caption": "31 Doom-03: key+camera excl rand", "image_path": "path_to_validation/000003.jpg", "action_path": "path_to_validation/key_camera_excl_1_action_rand_1.npy", "video_path": null, "num_inference_steps": 40, "height": 352, "width": 640, "num_frames": 77 } ] }