Competition

Evaluation Criteria

We have three devided datasets, training, validation, test. Submission are only evaludated the test dataset. We use the standard MPJPE, PA-MPJPE, V2V and PA-V2V metrics for evaluation:

• MPJPE: Mean Per-Joint Position Error (in mm) is the mean Euclidean distance between predicted and ground-truth 3D joints, evaluated on 24 SMPL body joints. We evaluate MPJPE with translation-only alignment (aligns the bodies at the pelvis joint)
• PA-MPJPE: MPJPE after Procrustes Alignment (in mm), solves for scale, translation and rotation
• V2V: Vertex-to-Vertex Error (in mm) is the mean Euclidean distance over all body vertices, computed between two SMPL meshes
• PA-V2V: V2V error after Procrustes Alignment (in mm)

Terms and Conditions

To participate in this challenge you need to agree with the following conditions: https://egobody.ethz.ch/

Submission Format

To submit your results to the leaderboard you must construct a submission json:

• pred_smpl.json - SMPL params prediction on the test set

JSON Submission Format

You need to put the recording_name (e.x., recording_20210907_S03_S04_01) as the key of first level, and the egocentric RGB frame name (e.x., '132754997786014666_frame_01442.jpg') as the key of the second level. The predicted SMPL params of each egocentric RGB frame serve as the third level (the params include the keys of 'gender', 'global_orient', 'betas', 'body_pose' ). Dimension or data format of each SMPL param: 'gender': 'female' / 'male' / 'neutral' 'global_orient': 1 * 3 'betas': 1 * 10 'body_pose': 1 * 69 The detailed format is as following: { "recording_20210907_S03_S04_01": {"132754997786014666_frame_01442.jpg": {"gender": "neutral", "global_orient": [[-1.84886465, 0.23738672, -2.49125209]], "betas": [[1.2807611, 0.89183897, 2.2324522, 1.2157959, 0.46208212, -0.11218458, 0.4492286, -0.35207522, 0.9982906, -0.8182647]], "body_pose": [[-0.08753302, -0.01012669, 0.02800051, ..., -0.91448868, 0.19536538, 0.91873705]], }, }, {"132754997786347863_frame_01443.jpg": {"gender": "neutral", "global_orient": [[-1.80268681, 0.22141867, -2.52025744]], "betas": [[1.2804753, 0.8915529, 2.2168348, 1.2592449, 0.4587577, -0.06185403, 0.4467768, -0.3691427, 1.0372409, -0.7852913]], "body_pose": [[-1.24743940e-01, -2.64450869e-02, -5.42001497e-04, ..., -9.19680595e-01, 1.91903051e-01, 9.27651146e-01]] }, }, ... "recording_20210911_S03_S08_01": { ... }, ... } Here you can find an example pred_smpl.json produced by evaluating SPIN on the EgoBody test set. Please submit your result in pred_smpl.json to CodaLab.