Add Lite-v2 (manually refined) config + README note; CVPR 2026 acceptance

README.md

@@ -25,6 +25,10 @@ configs:
   data_files:
   - split: test
     path: data/herbench_lite.parquet
+- config_name: lite_v2
+  data_files:
+  - split: test
+    path: data/herbench_lite_v2.parquet
 ---
 
 # HERBench: A Benchmark for Multi-Evidence Integration in Video Question Answering
@@ -38,11 +42,21 @@ configs:
 [![Project Page](https://img.shields.io/badge/Project-Page-blue)](https://gabrieleserussi.github.io/HERBench/)
 [![License](https://img.shields.io/badge/License-CC%20BY--NC--SA%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
 [![HF Dataset](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Dataset-yellow)](https://huggingface.co/datasets/DanBenAmi/HERBench)
+[![CVPR 2026](https://img.shields.io/badge/CVPR-2026-1f72b8.svg)](https://cvpr.thecvf.com/)
 
 *A challenging benchmark for evaluating multi-evidence integration capabilities of vision-language models*
 
+**🎉 HERBench has been accepted to CVPR 2026!**
+
 </div>
 
+> **🆕 New: Lite-v2 config.** We released a refreshed **`lite_v2`** version of the
+> Lite split (1,971 questions / 68 videos) in which 9 of the 12 tasks were
+> regenerated and went through **additional manual refinement** for higher
+> quality, while TSO, SVA, and RLPC are carried over unchanged from `lite`.
+> Load it with `load_dataset("DanBenAmi/HERBench", "lite_v2")`. The original
+> `lite` config remains available unchanged.
+
 ---
 
 ## 📋 Dataset Summary
@@ -55,12 +69,15 @@ configs:
 
 ### Key Statistics
 
-| Metric | Full Version | Lite Version |
-|--------|--------------|--------------|
-| 📊 **Total Questions** | 27,631 five-way multiple-choice | 2,000 questions (7.2%) |
-| 🎬 **Videos** | 335 unique videos | 68 unique videos (20.3%) |
-| ⏱️ **Avg. Video Length** | 424 seconds | 421 seconds |
-| 📁 **Total Size** | ~161 GB | ~35 GB |
+| Metric | Full Version | Lite Version | Lite-v2 (refined) |
+|--------|--------------|--------------|-------------------|
+| 📊 **Total Questions** | 27,631 five-way multiple-choice | 2,000 questions (7.2%) | 1,971 questions |
+| 🎬 **Videos** | 335 unique videos | 68 unique videos (20.3%) | 68 unique videos |
+| ⏱️ **Avg. Video Length** | 424 seconds | 421 seconds | 421 seconds |
+| 📁 **Total Size** | ~161 GB | ~35 GB | ~35 GB (same videos) |
+
+> **Lite-v2** uses the same 68 videos as Lite. 9 of the 12 tasks were regenerated
+> with additional manual refinement; TSO, SVA, and RLPC are identical to Lite.
 
 ### Why HERBench?
 
@@ -274,6 +291,7 @@ huggingface-cli download DanBenAmi/HERBench --repo-type dataset --local-dir HERB
 # Download LITE version only (2,000 questions, ~35 GB videos)
 huggingface-cli download DanBenAmi/HERBench \
     --include "data/herbench_lite.parquet" \
+    --include "data/herbench_lite_v2.parquet" \
     --include "data/*metadata.json" \
     --include "videos/videos.tar.part.00" \
     --include "videos/videos.tar.part.01" \
@@ -312,6 +330,10 @@ print(f"Video: {example['video_path']}")
 # Load LITE version - 2,000 questions (7.2% sample)
 dataset_lite = load_dataset("DanBenAmi/HERBench", "lite")
 print(f"Lite questions: {len(dataset_lite['test'])}")
+
+# Load LITE-v2 (refined) - 1,971 questions, same 68 videos
+dataset_lite_v2 = load_dataset("DanBenAmi/HERBench", "lite_v2")
+print(f"Lite-v2 questions: {len(dataset_lite_v2['test'])}")
 ```
 
 **Schema:** Each example contains:

data/herbench_annotations_lite_v2.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:247eeca9f76cf1745a5d853a111463f0bb1af4f64fa1d945df6c40c15abd88a1
+size 2595653

data/herbench_lite_v2.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23a7dc1d6baafc75ac9065cf13591e6dc1c7ef92f044bb9d2f00067b8f861c8c
+size 462396