Add script to generate embedding for dataset #2138
Open
+924
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
pkooij
added
dataset
Comment on lines
+201
to
+228
| if isinstance(img, torch.Tensor): | ||
| if img.ndim == 4: | ||
| # Shape: (T, C, H, W) where T=1 for single timestamp | ||
| img = img[0] # Now (C, H, W) | ||
| elif img.ndim == 3: | ||
| # Shape: (C, H, W) | ||
| pass | ||
| else: | ||
| raise ValueError( | ||
| f"Unexpected video frame shape {img.shape} for camera {cam_key}. " | ||
| f"Expected (T, C, H, W) or (C, H, W). Episode {ep_idx}, Frame {frame_idx}" | ||
| ) | ||
|
|
||
| # Convert to numpy: (C, H, W) float32 [0, 1] -> (H, W, C) uint8 [0, 255] | ||
| img_np = (img.permute(1, 2, 0).numpy() * 255).astype(np.uint8) | ||
| else: | ||
| img_np = np.array(img) | ||
| else: | ||
| # Load from image file | ||
| img = item[cam_key] | ||
| # Convert to numpy if needed | ||
| if isinstance(img, torch.Tensor): | ||
| if img.ndim == 3: | ||
| img_np = (img.permute(1, 2, 0).numpy() * 255).astype(np.uint8) | ||
| else: | ||
| raise ValueError(f"Unexpected image shape {img.shape} for camera {cam_key}") | ||
| else: | ||
| img_np = np.array(img) |
There was a problem hiding this comment.
Should we add testing for this part,
this logic could be buggy if we have image datasets instead of videos.
Better if we had tests to verify
| self.batch_size = batch_size | ||
| self.model_name = model_name | ||
| logger.info(f"Loading DinoV2 model: {model_name}") | ||
| self.model = torch.hub.load("facebookresearch/dinov2", model_name) # nosec B614 |
There was a problem hiding this comment.
Is there a reason why you didn't use AutoModel from transformers here also? We do it for the SAC encoder.
Something like:
self.model = AutoModel.from_pretrained("facebook/dinov2_base"):
| from lerobot.datasets.lerobot_dataset import LeRobotDataset | ||
|
|
||
|
|
||
| def cosine_similarity(a: np.ndarray, b: np.ndarray) -> float: |
Comment on lines
+275
to
+292
| for (chunk_idx, file_idx), frame_indices in tqdm(frames_by_file.items(), desc="Updating parquet files"): | ||
| parquet_path = output_dataset.root / DEFAULT_DATA_PATH.format( | ||
| chunk_index=chunk_idx, file_index=file_idx | ||
| ) | ||
|
|
||
| # Load the parquet file | ||
| df = pd.read_parquet(parquet_path) | ||
|
|
||
| # Add embedding columns | ||
| df["task_embedding"] = [all_task_embeddings[idx].tolist() for idx in frame_indices] | ||
|
|
||
| for cam_key in dataset.meta.camera_keys: | ||
| df[f"{cam_key}_embedding"] = [ | ||
| image_embeddings_dict[cam_key][idx].tolist() for idx in frame_indices | ||
| ] | ||
|
|
||
| # Save the updated parquet file | ||
| df.to_parquet(parquet_path, index=False) |
There was a problem hiding this comment.
For this chunk I suggest to do it in a more effiecient way either by using dataset_tools. I'll push a PR.
Reading a writing the parquet file to disk will be slow and result in memory explosion for huge datasets.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR introduces a way to generate image and text embeddings to a dataset to be more efficient when training a dataset for multiple epochs. For example for learning a general reward we contain a specific dataset with OXE to improve generalization. In order to nor recompute the image and text embeddings each time we finetune for OXE we can use this script to add the embeddings to the dataset. We can additionally remove the videos in the dataset to safe space.
Testing:
Both the generate and validate script were tested on this dataset: lerobot/utokyo_xarm_bimanual. The generated dataset can be found here: pepijn223/utokyo_xarm_bimanual_embeddings.