Add paper, code, and project page links; add image-text-to-text task category

README.md

@@ -1,12 +1,13 @@
 ---
-license: cc-by-4.0
-task_categories:
-- visual-question-answering
 language:
 - en
-pretty_name: BedBench
+license: cc-by-4.0
 size_categories:
 - 1K<n<10K
+task_categories:
+- visual-question-answering
+- image-text-to-text
+pretty_name: BedBench
 configs:
 - config_name: T1_uncolor
   data_files:
@@ -151,11 +152,12 @@ configs:
   - split: train
     path: data/T13_valid_color/train-*.parquet
 ---
+
 # BedBench
 
 Evaluation data, model outputs, and probing features for the paper
-*Where Do Vision-Language Models Fail at 3D Reasoning?* Every task is posed on rendered views of 100 procedurally generated voxel objects (T1-T12) or 100
-multi-object scenes (T13). Code (generator, evaluation harness, probing, fine-tuning) is released separately.
+[*Where Do Vision-Language Models Fail at 3D Reasoning?*](https://huggingface.co/papers/2610.39639). Every task is posed on rendered views of 100 procedurally generated voxel objects (T1-T12) or 100
+multi-object scenes (T13). Code (generator, evaluation harness, probing, fine-tuning) is released separately at [GitHub](https://github.com/leduckhai/BedBench); see the [project page](https://leduckhai.github.io/BedBench) for more details.
 
 ## Configurations
 
@@ -201,4 +203,4 @@ Main developer: [Nguyen Kim Hai Bui](https://whynotkimhari.github.io) ([GitHub](
 
 ## License
 
-CC BY 4.0.
+CC BY 4.0.