--- license: apache-2.0 task_categories: - image-segmentation - depth-estimation - image-to-image - image-feature-extraction - image-text-to-text language: - en tags: - vlm - multimodal - visual-reasoning - chain-of-thought dataset_info: - config_name: part1 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 446733439.675 num_examples: 44475 download_size: 17934592717 dataset_size: 446733439.675 - config_name: part10 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 17744642822.628 num_examples: 44474 download_size: 17769519559 dataset_size: 17744642822.628 - config_name: part11 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 17749640543.646 num_examples: 44474 download_size: 40708430498 dataset_size: 17749640543.646 - config_name: part12 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 16126888186.176 num_examples: 44474 download_size: 13512012899 dataset_size: 16126888186.176 - config_name: part13 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 12426535004.444 num_examples: 44474 download_size: 17473903640 dataset_size: 12426535004.444 - config_name: part14 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 18713928974.662 num_examples: 44474 download_size: 15650533270 dataset_size: 18713928974.662 - config_name: part15 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 5895768695.898 num_examples: 44474 download_size: 4400917224 dataset_size: 5895768695.898 - config_name: part16 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 6486479074.432 num_examples: 44474 download_size: 6271251512 dataset_size: 6486479074.432 - config_name: part17 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 6971738527.692 num_examples: 44474 download_size: 6469686725 dataset_size: 6971738527.692 - config_name: part18 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 7515615259.548 num_examples: 44474 download_size: 7563582116 dataset_size: 7515615259.548 - config_name: part19 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 7607522167.776 num_examples: 44474 download_size: 7377834021 dataset_size: 7607522167.776 - config_name: part2 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 33365826763.55 num_examples: 44475 download_size: 33245002109 dataset_size: 33365826763.55 - config_name: part20 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 7179023719.416 num_examples: 44474 download_size: 6145041547 dataset_size: 7179023719.416 - config_name: part3 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 24289174922.225 num_examples: 44475 download_size: 15536031938 dataset_size: 24289174922.225 - config_name: part4 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 6699346145.65 num_examples: 44475 download_size: 6646535744 dataset_size: 6699346145.65 - config_name: part5 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 6651458829.325 num_examples: 44475 download_size: 4896128944 dataset_size: 6651458829.325 - config_name: part6 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 18039837306.175 num_examples: 44475 download_size: 17786918533 dataset_size: 18039837306.175 - config_name: part7 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 17852309246.925 num_examples: 44475 download_size: 17774641127 dataset_size: 17852309246.925 - config_name: part8 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 17873005196.425 num_examples: 44475 download_size: 16131787376 dataset_size: 17873005196.425 - config_name: part9 features: - name: id dtype: string - name: conversations list: - name: from dtype: string - name: value dtype: string - name: image_relpath dtype: string - name: image dtype: image splits: - name: train num_bytes: 34314021153.775 num_examples: 44475 download_size: 41828939793 dataset_size: 34314021153.775 configs: - config_name: part1 data_files: - split: train path: part1/train-* - config_name: part10 data_files: - split: train path: part10/train-* - config_name: part11 data_files: - split: train path: part11/train-* - config_name: part12 data_files: - split: train path: part12/train-* - config_name: part13 data_files: - split: train path: part13/train-* - config_name: part14 data_files: - split: train path: part14/train-* - config_name: part15 data_files: - split: train path: part15/train-* - config_name: part16 data_files: - split: train path: part16/train-* - config_name: part17 data_files: - split: train path: part17/train-* - config_name: part18 data_files: - split: train path: part18/train-* - config_name: part19 data_files: - split: train path: part19/train-* - config_name: part2 data_files: - split: train path: part2/train-* - config_name: part20 data_files: - split: train path: part20/train-* - config_name: part3 data_files: - split: train path: part3/train-* - config_name: part4 data_files: - split: train path: part4/train-* - config_name: part5 data_files: - split: train path: part5/train-* - config_name: part6 data_files: - split: train path: part6/train-* - config_name: part7 data_files: - split: train path: part7/train-* - config_name: part8 data_files: - split: train path: part8/train-* - config_name: part9 data_files: - split: train path: part9/train-* --- # CoVT Dataset This repository contains the training data for **Chain-of-Visual-Thought (CoVT)**, a framework proposed in the paper "[Chain-of-Visual-Thought: Teaching VLMs to See and Think Better with Continuous Visual Tokens](https://huggingface.co/papers/2511.19418)". CoVT enables Vision-Language Models (VLMs) to reason not only in words but also through **continuous visual tokens**—compact latent representations that encode rich perceptual cues. This framework addresses the limitations of current VLMs in capturing dense visual information across spatial dimensions, which is essential for perceptual understanding, spatial reasoning, and geometric awareness. - **Paper:** [https://huggingface.co/papers/2511.19418](https://huggingface.co/papers/2511.19418) - **Project Page:** https://wakalsprojectpage.github.io/comt-website - **Code:** https://github.com/Wakals/CoMT ## Dataset Composition The CoVT dataset is designed to provide dense supervision signals for training VLMs. It utilizes some subsets of LLaVA-OneVision and merges the filtered TallyQA dataset and ADE20K-Depth from Aurora. During training, a VLM equipped with CoVT autoregressively predicts these visual tokens to reconstruct dense supervision signals such as depth, segmentation, edges, and DINO features. ![Data Image](https://github.com/Wakals/CoMT/raw/main/assets/data.png) ## Sample Usage To load a specific part of the CoVT dataset, you can use the Hugging Face `datasets` library: ```python from datasets import load_dataset # Load a specific configuration (e.g., 'part1') dataset = load_dataset("Wakals/CoVT-Dataset", "part1") # Access the training split train_data = dataset["train"] # Print the first example print(train_data[0]) # Example of accessing image and conversations from a sample # first_example = train_data[0] # image_obj = first_example["image"] # This will be a PIL Image object # print(f"Image format: {image_obj.format}, size: {image_obj.size}") # print(f"Conversations: {first_example['conversations']}") ```