Method process_scene

benchmark/methods/lingbot_map.py:258–293

Process a scene with LingbotMap inference.

(self, gt_artifact)

        return rgb_list, depth_list, pose_list, intrinsics_list, confidence_list

    def process_scene(self, gt_artifact) -> Dict[str, Any]:
        """Process a scene with LingbotMap inference."""
        loader = BSSLoader(gt_artifact, resize_context=self.resize_context)
        input_rgb_list = loader.load_rgb_list()
        self.logger.info(f"Image size for processing: {loader.get_processing_dimensions()} (HxW)")

        print(f"  → Processing {len(input_rgb_list)} frames with LingbotMap (mode: {self.mode})")

        # Prepare and run inference
        images = self._prepare_images(input_rgb_list)
        image_shape = images.shape[-2:]  # (H, W)
        predictions = self._run_inference(images)

        # Convert outputs
        rgb_list, depth_list, pose_list, intrinsics_list, confidence_list = \
            self._process_outputs(predictions, image_shape)

        if len(depth_list) != len(input_rgb_list):
            print(f"  → WARNING: Output frames ({len(depth_list)}) != input frames ({len(input_rgb_list)})")

        # Assemble results
        print(f"  → Assembling {len(rgb_list)} frames in standard format")
        frame_results = {
            'rgb': rgb_list,
            'depth': depth_list,
            'pose': pose_list,
            'intrinsics': intrinsics_list,
        }

        if confidence_list:
            frame_results['confidence'] = confidence_list

        return {
            'frame': frame_results,
            'global': {},
        }