docs: add probe model card and feature-layer clarification

README.md

@@ -0,0 +1,76 @@
+---
+language:
+  - en
+tags:
+  - spatial-intelligence
+  - frozen-feature-probing
+  - semantic-tagging
+  - instance-grouping
+  - 3d-geometry
+---
+
+# Probing-VLM-VGM Checkpoints
+
+[Paper](https://arxiv.org/abs/2605.28132) | [GitHub](https://github.com/om-ai-lab/Probing-VLM-VGM)
+
+Official trained probe checkpoints for **Which Pretraining Paradigm Better Serves Spatial Intelligence? An Empirical Comparison of Vision-Language and Video Generation Models**.
+
+## About
+
+This repository provides **54 trained probes covering 18 foundation models and three tasks**: semantic tagging and instance grouping on ScanNet, and 3D geometry prediction on DL3DV. The probes read frozen VLM or VGM features to assess their semantic, object-level, and geometric information.
+
+These checkpoints contain the **probe weights**, not the frozen foundation-model weights. They are compact Lightning checkpoints containing the model state dictionary and Lightning version, without optimizer or other resumable training state. Datasets and extracted features are not included.
+
+For installation, data preparation, feature extraction, and download/evaluation commands, see the [GitHub README](https://github.com/om-ai-lab/Probing-VLM-VGM#-training-and-evaluation).
+
+## Repository Layout
+
+```text
+.
+|-- README.md
+|-- manifest.json
+|-- semantic_tagging/
+|   `-- <model>/
+|       |-- probe.ckpt
+|       |-- probe_config.json
+|       |-- metadata.json
+|       |-- clip_class_embeds.npy
+|       `-- class_names.json
+|-- instance_grouping/
+|   `-- <model>/
+|       |-- probe.ckpt
+|       |-- probe_config.json
+|       `-- metadata.json
+`-- dl3dv_geometry/
+    `-- <model>/
+        |-- probe.ckpt
+        |-- probe_config.json
+        `-- metadata.json
+```
+
+Each task directory contains one subdirectory per model, such as `qwen3-vl-8b` or `wan-t2v-14b`. `probe_config.json` specifies the probe architecture; `metadata.json` records the checkpoint's actual feature settings and provenance. `manifest.json` indexes all checkpoints and their SHA-256 checksums. The semantic-tagging directories additionally include CLIP class-name embeddings and the ordered class vocabulary.
+
+## Feature-Layer Clarification
+
+The paper's implementation details specify **layer 21 for Qwen2.5-VL** and **layer 22 for Qwen3-VL** in the main 3D experiments. The following uploaded checkpoints were instead trained with **layer 18** features:
+
+| Model | Layer stated in the paper's main 3D settings | Uploaded checkpoints' feature layer |
+| --- | --- | --- |
+| Qwen2.5-VL-3B | 21 | 18 |
+| Qwen2.5-VL-7B | 21 | 18 |
+| Qwen3-VL-2B | 22 | 18 |
+
+For these three models, the uploaded checkpoints use layer 18 across **all three tasks**. Please follow the settings in each checkpoint's `metadata.json` rather than assuming that the release configuration defaults apply: extract features with `--output-layers 18` and evaluate with `feat_postfix=_layer18`, together with the packaged probe architecture. Features from different layers are not interchangeable, even when their tensor shapes match.
+
+Our spot checks of the uploaded probes reproduce the paper's reported metrics under matching evaluation settings. In particular, the uploaded **Qwen2.5-VL-7B layer-18 DL3DV probe** achieves **0.234 P-map Err., 0.127 depth AbsRel, and 0.310 camera AUC@30**, matching the paper's displayed values. This verifies the tested checkpoint against the reported results; it is not a claim that every layer variant has been evaluated or that different layers give identical results. Exact reproduction also depends on evaluation sampling and dependency versions.
+
+## Citation
+
+```bibtex
+@article{shen2026probingvlmvgm,
+  title={Which Pretraining Paradigm Better Serves Spatial Intelligence? An Empirical Comparison of Vision-Language and Video Generation Models},
+  author={Shen, Haozhan and Zhao, Tiancheng and Zhao, Kangjia and Yin, Jianwei},
+  journal={arXiv preprint arXiv:2605.28132},
+  year={2026}
+}
+```