{
    "data": [
        {
            "caption": "00",
            "image_path": "path_to_validation/00.png",
            "ref_video": "path_to_validation/00.mp4",
            "action_path": "path_to_validation/00.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "01",
            "image_path": "path_to_validation/01.png",
            "ref_video": "path_to_validation/01.mp4",
            "action_path": "path_to_validation/01.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "02",
            "image_path": "path_to_validation/02.png",
            "ref_video": "path_to_validation/02.mp4",
            "action_path": "path_to_validation/02.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "03",
            "image_path": "path_to_validation/03.png",
            "ref_video": "path_to_validation/03.mp4",
            "action_path": "path_to_validation/03.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "04",
            "image_path": "path_to_validation/04.png",
            "ref_video": "path_to_validation/04.mp4",
            "action_path": "path_to_validation/04.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "05",
            "image_path": "path_to_validation/05.png",
            "ref_video": "path_to_validation/05.mp4",
            "action_path": "path_to_validation/05.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "06",
            "image_path": "path_to_validation/06.png",
            "ref_video": "path_to_validation/06.mp4",
            "action_path": "path_to_validation/06.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "07",
            "image_path": "path_to_validation/07.png",
            "ref_video": "path_to_validation/07.mp4",
            "action_path": "path_to_validation/07.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "08",
            "image_path": "path_to_validation/08.png",
            "ref_video": "path_to_validation/08.mp4",
            "action_path": "path_to_validation/08.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "09",
            "image_path": "path_to_validation/09.png",
            "ref_video": "path_to_validation/09.mp4",
            "action_path": "path_to_validation/09.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "10",
            "image_path": "path_to_validation/10.png",
            "ref_video": "path_to_validation/10.mp4",
            "action_path": "path_to_validation/10.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "11",
            "image_path": "path_to_validation/11.png",
            "ref_video": "path_to_validation/11.mp4",
            "action_path": "path_to_validation/11.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "12",
            "image_path": "path_to_validation/12.png",
            "ref_video": "path_to_validation/12.mp4",
            "action_path": "path_to_validation/12.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "13",
            "image_path": "path_to_validation/13.png",
            "ref_video": "path_to_validation/13.mp4",
            "action_path": "path_to_validation/13.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "14",
            "image_path": "path_to_validation/14.png",
            "ref_video": "path_to_validation/14.mp4",
            "action_path": "path_to_validation/14.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "15",
            "image_path": "path_to_validation/15.png",
            "ref_video": "path_to_validation/15.mp4",
            "action_path": "path_to_validation/15.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "16",
            "image_path": "path_to_validation/16.png",
            "ref_video": "path_to_validation/16.mp4",
            "action_path": "path_to_validation/16.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "17",
            "image_path": "path_to_validation/17.png",
            "ref_video": "path_to_validation/17.mp4",
            "action_path": "path_to_validation/17.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "18",
            "image_path": "path_to_validation/18.png",
            "ref_video": "path_to_validation/18.mp4",
            "action_path": "path_to_validation/18.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "19",
            "image_path": "path_to_validation/19.png",
            "ref_video": "path_to_validation/19.mp4",
            "action_path": "path_to_validation/19.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "20",
            "image_path": "path_to_validation/20.png",
            "ref_video": "path_to_validation/20.mp4",
            "action_path": "path_to_validation/20.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "21",
            "image_path": "path_to_validation/21.png",
            "ref_video": "path_to_validation/21.mp4",
            "action_path": "path_to_validation/21.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "22",
            "image_path": "path_to_validation/22.png",
            "ref_video": "path_to_validation/22.mp4",
            "action_path": "path_to_validation/22.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "23",
            "image_path": "path_to_validation/23.png",
            "ref_video": "path_to_validation/23.mp4",
            "action_path": "path_to_validation/23.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "24",
            "image_path": "path_to_validation/24.png",
            "ref_video": "path_to_validation/24.mp4",
            "action_path": "path_to_validation/24.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "25",
            "image_path": "path_to_validation/25.png",
            "ref_video": "path_to_validation/25.mp4",
            "action_path": "path_to_validation/25.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "26",
            "image_path": "path_to_validation/26.png",
            "ref_video": "path_to_validation/26.mp4",
            "action_path": "path_to_validation/26.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "27",
            "image_path": "path_to_validation/27.png",
            "ref_video": "path_to_validation/27.mp4",
            "action_path": "path_to_validation/27.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "28",
            "image_path": "path_to_validation/28.png",
            "ref_video": "path_to_validation/28.mp4",
            "action_path": "path_to_validation/28.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "29",
            "image_path": "path_to_validation/29.png",
            "ref_video": "path_to_validation/29.mp4",
            "action_path": "path_to_validation/29.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "30",
            "image_path": "path_to_validation/30.png",
            "ref_video": "path_to_validation/30.mp4",
            "action_path": "path_to_validation/30.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        },
        {
            "caption": "31",
            "image_path": "path_to_validation/31.png",
            "ref_video": "path_to_validation/31.mp4",
            "action_path": "path_to_validation/31.npy",
            "video_path": null,
            "num_inference_steps": 4,
            "height": 352,
            "width": 640,
            "num_frames": 81
        }
    ]
}