Add dataset statistics, teaser, and release plan

README.md

@@ -0,0 +1,40 @@
+---
+pretty_name: APM-Bench
+language:
+  - en
+tags:
+  - egocentric-video
+  - streaming-video
+  - persistent-memory
+---
+
+# APM-Bench
+
+> **APM-Bench: Benchmarking cross-Session Persistent Memory for Real-World Egocentric Streaming Video Assistants**
+
+[Paper](https://arxiv.org/pdf/2609.37559) · [Project page](https://jianguo-huang11.github.io/APM-Bench/) · [Code](https://github.com/Jianguo-Huang11/APM-Bench)
+
+![APM-Bench teaser: activity-related sessions separated by real-world time gaps.](assets/teaser.png)
+
+APM-Bench groups egocentric video into activity-related trajectories that span separate interactions. Its tasks test whether an assistant can use earlier experience while answering questions, perceiving the current scene, and deciding when to respond.
+
+## Dataset statistics
+
+| Video source | Trajectories | Sessions | Candidates |
+| --- | ---: | ---: | ---: |
+| EgoLife | 76 | 472 | 1,896 |
+| HD-EPIC | 28 | 77 | 823 |
+| **Total** | **104** | **549** | **2,719** |
+
+| Capability | Candidates |
+| --- | ---: |
+| Cross-session understanding | 887 |
+| Real-time perception | 716 |
+| Adaptive response | 1,116 |
+
+The benchmark covers 12 tasks and 5,371 evaluation probes. The counts above describe the complete benchmark; this repository currently contains its 549 session videos.
+
+## Release plan
+
+- [x] Initialize the [dataset repository](https://huggingface.co/datasets/Jianguo-Huang11/APM-Bench) and [code repository](https://github.com/Jianguo-Huang11/APM-Bench).
+- [ ] Release the dataset and code before November 2026. We are preparing the annotations and evaluation code now.