Datasets:
Download meta/depth_encoding.json from alphabot2/12-Aug_pick_depth: direct link, hf CLI and curl.
- Browser
- Download file 2.61 kB
-
https://huggingface.co/datasets/alphabot2/12-Aug_pick_depth/resolve/main/meta/depth_encoding.json
- Command line
-
hf download hf://datasets/alphabot2/12-Aug_pick_depth/meta/depth_encoding.json
-
curl -L -o depth_encoding.json https://huggingface.co/datasets/alphabot2/12-Aug_pick_depth/resolve/main/meta/depth_encoding.json
2.61 kB
| { | |
| "source_topic_suffix": "aligned_depth_to_color/image_raw/compressedDepth", | |
| "source_format": "16UC1; compressedDepth", | |
| "units": "millimeters", | |
| "packing": "R=depth_mm>>8, G=depth_mm&0xFF, B=0", | |
| "decode": "depth_mm = (rgb[...,0].astype('uint16') << 8) | rgb[...,1]", | |
| "decode_from_lerobot": "LeRobotDataset returns these keys as float32 CHW in [0,1]; recover the bytes with rgb = (t.permute(1,2,0).numpy()*255).round().astype('uint8') before unpacking. See depth_from_lerobot_frame() in convert_aibot2_rosbags_depth.py.", | |
| "invalid_value": 0, | |
| "max_representable_mm": 65535, | |
| "storage_dtype": "image", | |
| "storage_note": "Stored as lossless PNG. Do NOT re-encode these keys as video: lossy compression destroys the byte packing.", | |
| "resize_interpolation": "nearest", | |
| "keys": [ | |
| "observation.images.camera_front_depth", | |
| "observation.images.camera_left_depth", | |
| "observation.images.camera_right_depth" | |
| ], | |
| "viz": { | |
| "purpose": "Human-viewable preview of the matching packed depth key.", | |
| "lossy": true, | |
| "do_not_train_on": true, | |
| "derived_from": "the '<camera>_depth' key in the same dataset", | |
| "colormap": "OpenCV TURBO (falls back to JET), applied to depth clipped to range_mm", | |
| "invalid_pixels": "depth_mm == 0 renders black. Dropout within fill_holes_radius_px of real data is filled from the nearest valid pixel, and valid islands smaller than min_island_px are discarded as noise -- both in the preview only. The packed key keeps every zero and every sample.", | |
| "storage_dtype": "image", | |
| "range_note": "One window per depth camera, constant across the whole dataset, so a given colour means the same distance in every frame. Near end = blue, far = red.", | |
| "keys": [ | |
| "observation.images.camera_front_depth_viz", | |
| "observation.images.camera_left_depth_viz", | |
| "observation.images.camera_right_depth_viz" | |
| ], | |
| "range_mm": { | |
| "observation.images.camera_front_depth_viz": { | |
| "min_mm": 214.0, | |
| "max_mm": 1224.0 | |
| }, | |
| "observation.images.camera_left_depth_viz": { | |
| "min_mm": 81.0, | |
| "max_mm": 2381.0 | |
| }, | |
| "observation.images.camera_right_depth_viz": { | |
| "min_mm": 91.0, | |
| "max_mm": 1494.0 | |
| } | |
| }, | |
| "scale": "log", | |
| "fill_holes_radius_px": 4, | |
| "min_island_px": 256, | |
| "median_smooth_px": 3, | |
| "temporal_fill_frames": 5, | |
| "stream_keys": [ | |
| "observation.images.camera_front_depth_stream", | |
| "observation.images.camera_left_depth_stream", | |
| "observation.images.camera_right_depth_stream" | |
| ], | |
| "stream_storage_dtype": "video" | |
| } | |
| } |