VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement
Paper • 2607.00446 • Published • 16
How to use VideoSearchR1/didemo-stage1 with Transformers:
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM
processor = AutoProcessor.from_pretrained("VideoSearchR1/didemo-stage1")
model = AutoModelForMultimodalLM.from_pretrained("VideoSearchR1/didemo-stage1")This is the Stage 1 VideoSearch-R1 checkpoint trained for DiDeMo, presented in VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement.
Use with the VideoSearch-R1 codebase:
bash scripts/data_construct/download_preextracted_data.bash didemo
EVAL_GPUS=0 bash scripts/inference/inference.bash didemo --checkpoint VideoSearchR1/didemo-stage1
@inproceedings{lee2026videosearchr1,
title = {VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement},
author = {Lee, Seohyun and Choi, Seoung and Ko, Dohwan and Kim, Jongha and Kim, Hyunwoo J.},
booktitle = {European Conference on Computer Vision (ECCV)},
year = {2026}
}
Base model
Qwen/Qwen3-VL-4B-Instruct