Add paper link, pipeline tag, and repository info

README.md

@@ -1,7 +1,8 @@
 ---
+base_model: Qwen/Qwen2.5-VL-3B-Instruct
 library_name: transformers
+pipeline_tag: video-text-to-text
 license: other
-base_model: Qwen/Qwen2.5-VL-3B-Instruct
 tags:
 - llama-factory
 - full
@@ -13,19 +14,27 @@ model-index:
 
 # CaMo-3B
 
-This model is a fine-tuned version of [Qwen/Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) on the camerabench_10K and the spld datasets.
+This model is a fine-tuned version of [Qwen/Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) on the **CameraBench** and the **SpatialLadder** datasets.
+
+It was introduced in the paper [CaMo: Camera Motion Grounded Evaluation and Training for Vision-Language Models](https://arxiv.org/abs/2605.20165).
+
+- **Code:** [GitHub Repository](https://github.com/hsiangwei0903/CaMo)
 
 ## Model description
 
-More information needed
+CaMo is a camera motion grounded Vision-Language Model (VLM) designed to address the gap in spatial cognition among existing VLMs. While state-of-the-art models often perform well on static spatial QA benchmarks, they frequently lack an understanding of camera motion—a key component of spatial intelligence.
+
+CaMo is trained to generate explicit spatial narratives that capture both scene semantics and camera motion. It achieves consistent performance across the proposed Spatial Narrative Score (SNS) evaluation and direct spatial question answering tasks.
 
 ## Intended uses & limitations
 
-More information needed
+The model is intended for research in 3D spatial understanding, video analysis, and robotics, where understanding camera trajectories (e.g., panning, tilting, zooming) is crucial.
 
 ## Training and evaluation data
 
-More information needed
+The model was fine-tuned on:
+- **SpatialLadder-26k**: A dataset for fine-grained spatial understanding.
+- **CameraBench**: A benchmark and training set specifically targeting camera motion understanding.
 
 ## Training procedure
 
@@ -41,18 +50,14 @@ The following hyperparameters were used during training:
 - gradient_accumulation_steps: 2
 - total_train_batch_size: 128
 - total_eval_batch_size: 64
-- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
 - num_epochs: 1
 
-### Training results
-
-
-
 ### Framework versions
 
 - Transformers 4.57.1
 - Pytorch 2.5.1+cu124
 - Datasets 4.0.0
-- Tokenizers 0.22.1
+- Tokenizers 0.22.1