VideoChat-R1
Collection
VideoChat-R1: Enhancing Spatio-Temporal Perception via Reinforcement Fine-Tuning
β’
4 items
β’
Updated
β’
8
[π GitHub]
[π Tech Report]
We provide a simple installation example below:
pip install transformers
Using qwen_vl_utils in https://github.com/OpenGVLab/VideoChat-R1/blob/main/Videochat-R1.5/src_eval/my_vision_process.py
Then you could use our model:
from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
model_path = "OpenGVLab/VideoChat-R1_5"
# default: Load the model on the available device(s)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
model_path, torch_dtype="auto", device_map="auto",
attn_implementation="flash_attention_2"
)
# default processer
processor = AutoProcessor.from_pretrained(model_path)
video_path = "your_video.mp4"
question = "your_qa.mp4"
num_percptions = 3
QA_THINK_GLUE = """Answer the question: "[QUESTION]" according to the content of the video.
Output your think process within the <think> </think> tags.
Then, provide your answer within the <answer> </answer> tags, output the corresponding letter of the option. At the same time, in the <glue> </glue> tags, present the precise time period in seconds of the video clips on which you base your answer to this question in the format of [(s1, e1), (s2, e2), ...]. For example: <think>...</think><answer>A</answer><glue>[(5.2, 10.4)]</glue>.
"""
QA_THINK = """Answer the question: "[QUESTION]" according to the content of the video.
Output your think process within the <think> </think> tags.
Then, provide your answer within the <answer> </answer> tags, output the corresponding letter of the option. For example: <think>...</think><answer>A</answer><glue>[(5.2, 10.4)]</glue>.
"""
def inference(video_path, prompt, model, processor, max_new_tokens=2048, device="cuda:0", client = None, pred_glue=None):
messages = [
{"role": "user", "content": [
{"type": "video",
"video": video_path,
'key_time':pred_glue,
"total_pixels": 128*12 * 28 * 28,
"min_pixels": 128 * 28 * 28,
},
{"type": "text", "text": prompt},
]
},
]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs, video_kwargs = process_vision_info(messages, return_video_kwargs=True, client = client)
fps_inputs = video_kwargs['fps']
inputs = processor(text=[text], images=image_inputs, videos=video_inputs, fps=fps_inputs, padding=True, return_tensors="pt")
inputs = inputs.to(device)
with torch.no_grad():
output_ids = model.generate(**inputs, max_new_tokens=max_new_tokens, use_cache=True)
generated_ids = [output_ids[i][len(inputs.input_ids[i]):] for i in range(len(output_ids))]
output_text = processor.batch_decode(generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
return output_text[0]
for percption in range(num_percptions):
if percption == num_percptions - 1:
example_prompt = QA_THINK.replace("[QUESTION]", item["problem"]["question"])
else:
example_prompt = QA_THINK_GLUE.replace("[QUESTION]", item["problem"]["question"])
ans = inference(video_path, example_prompt, model, processor, device=device, client=client, pred_glue=pred_glue)
pattern_glue = r'<glue>(.*?)</glue>'
match_glue = re.search(pattern_glue, ans, re.DOTALL)
# print(f'ann:{ans}')
answers.append(ans)
pred_glue = None
try:
if match_glue:
glue = match_glue.group(1)
pred_glue = ast.literal_eval(glue)
except Exception as e:
pred_glue = None
print(ans)
If you find this project useful in your research, please consider cite:
@article{li2025videochatr1,
title={VideoChat-R1: Enhancing Spatio-Temporal
Perception via Reinforcement Fine-Tuning},
author={Li, Xinhao and Yan, Ziang and Meng, Desen and Dong, Lu and Zeng, Xiangyu and He, Yinan and Wang, Yali and Qiao, Yu and Wang, Yi and Wang, Limin},
journal={arXiv preprint arXiv:2504.06958},
year={2025}
}
@article{yan2025videochatr15,
title={VideoChat-R1.5: Visual Test-Time Scaling to Reinforce Multimodal Reasoning by Iterative Perception},
author={Yan, Ziang and Li, Xinhao and He, Yinan and Zhengrong Yue and Zeng, Xiangyu and Wang, Yali and Qiao, Yu and Wang, Limin and Wang, Yi},
journal={arXiv preprint arXiv:2509.21100},
year={2025}
}