Add MindView project overview and release documentation

README.md

@@ -0,0 +1,46 @@
+---
+language:
+- en
+tags:
+- mindview
+- training-data
+- multi-view
+- spatial-intelligence
+- image-to-image
+---
+
+# MindView-Train
+
+**MindView-Train** is the training-data repository for *MindView: Benchmarking Generative Multi-View Spatial Intelligence*. The training workflow uses language queries and observed views to supervise generation of a target view from the same indoor scene.
+
+**Release status:** This repository currently contains documentation only. Training JSONL files and image assets have not been uploaded. Release packaging and access instructions are being prepared.
+
+## Training-data specification
+
+- **20,000 training examples** in the frozen six-view training specification.
+- **Six ordered context RGB images**, a language query, and **one target RGB image** per example.
+- Coverage of the project's seven target-view generation tasks.
+- Source scene resources from ScanNet, ScanNet++, and ARKitScenes.
+
+The training JSONL follows `mindview_bagel_single_shot_v1`: `shot` is `6`, `contexts["6"]` contains six relative RGB paths, and `target_image` identifies the separate supervision image. Paths resolve against the configured source-dataset roots.
+
+The training validation protocol requires unique record identities and source lineages, preserves context order, keeps the anchor first when the task requires one, and excludes the target from its context images. It also rejects scene identities shared with the frozen benchmark. This train–evaluation separation must be preserved when preparing downstream splits.
+
+## Related resources
+
+Collection: [MindView](https://huggingface.co/collections/wangyanbo/mindview-6acf6b7a4a2e25efd51989f5).
+
+| Resource | Purpose |
+|---|---|
+| [MindView-Train](https://huggingface.co/datasets/wangyanbo/MindView-Train) | Six-view supervised training data |
+| [MindView-Bench](https://huggingface.co/datasets/wangyanbo/MindView-Bench) | Frozen evaluation data |
+| [MindView-BAGEL-SFT](https://huggingface.co/wangyanbo/MindView-BAGEL-SFT) | Fine-tuned BAGEL EMA checkpoint |
+| [GitHub project repository](https://github.com/Yanbo-23/MindView) | Project code and usage documentation |
+
+Code and usage documentation are being prepared for publication at the project repository. Loading instructions, asset manifests, and checksums will be documented with the data release.
+
+## Access and terms
+
+Resource access and redistribution terms are being finalized. Source datasets retain their own terms; the code repository's license does not establish a license for the training-data assets. Please refer to the release terms before using or redistributing the eventual files.
+
+For project questions, contact [Yihao Liu](mailto:liuyihao@pjlab.org.cn) or [Jiwen Lu](mailto:lujiwen@tsinghua.edu.cn).