VincentNi/EmbodiedVideoRL_baseline_put_object_cabinet

第一轮（预NFT）Wan2.2 TI2V基线rollouts：put_object_cabinet任务。这些视频是在进行任何NFT微调之前生成的（NFT步骤0），用于作为参考分布来衡量NFT改进的效果。数据集包含10个场景，每个场景有16个rollout，总共160个视频。视频的生成配置包括模型（Wan2.2-TI2V-5B + merged_vidar_lora.pt）、分辨率（640 x 736）、帧率（121帧，24fps，约5.04秒）、采样方法（SDE，eta=1.0，50步，shift=5.0，guide_scale=5.0）等详细信息。

First-round (pre-NFT) Wan2.2 TI2V baseline rollouts: put_object_cabinet task. Baseline video rollouts generated before any NFT fine-tuning (NFT step 0) for the RoboTwin `put_object_cabinet` task. Used as a reference distribution for measuring NFT improvement. The dataset includes 10 scenes, with 16 rollouts per scene, totaling 160 videos. The generation configuration includes model (Wan2.2-TI2V-5B + merged_vidar_lora.pt), resolution (640 x 736), frame rate (121 frames, 24fps, ~5.04s), sampling method (SDE, eta=1.0, 50 steps, shift=5.0, guide_scale=5.0), and other details.