fix(global-recon-model): Camera variables in init stage (+1 more)

[andrewwhitecdw]

Small fixes in global_recon/models/global_recon_model.py:

Camera variables in init stage are optimized but never applied

Fix: Replace:

        if self.flag_opt_cam and opt_meta['stage'] != 'init':

with:

        if self.flag_opt_cam:

scale is None but indexed when predictor uses joint positions

Fix: Replace:

            if self.mt_model.traj_predictor is not None and self.mt_model.traj_predictor.in_joint_pos_only:
                batch['shape'] = pose_dict['smpl_beta'][exist_fr].unsqueeze(0)
                batch['scale'] = pose_dict['scale'][exist_fr].unsqueeze(0)

with:

            if self.mt_model.traj_predictor is not None and self.mt_model.traj_predictor.in_joint_pos_only:
                batch['shape'] = pose_dict['smpl_beta'][exist_fr].unsqueeze(0)
                batch['scale'] = pose_dict['scale'][exist_fr].unsqueeze(0) if pose_dict['scale'] is not None else None

Files changed

• global_recon/models/global_recon_model.py