VincentNi/wan22-rollout-put-object-cabinet-step0
收藏资源简介:
Wan2.2 TI2V Step-0 Rollout — RoboTwin `put_object_cabinet`数据集包含160个视频(10个场景×16个样本),由Wan2.2-TI2V-5B模型和合并的Vidar LoRA生成,用于RoboTwin `put_object_cabinet`任务。这些是用于评估奖励模型行为和种子RL微调的预NFT训练(step-0)基线样本。数据集的具体配置包括基础模型、LoRA、采样器、步骤、位移、CFG比例、帧数和种子等详细信息。数据集的布局和文件命名规则也在README中进行了说明。
The Wan2.2 TI2V Step-0 Rollout — RoboTwin `put_object_cabinet` dataset contains 160 videos (10 scenes × 16 samples) generated by the Wan2.2-TI2V-5B model and merged Vidar LoRA for the RoboTwin `put_object_cabinet` task. These are the pre-NFT-training (step-0) baseline samples used to evaluate reward-model behaviour and seed RL fine-tuning. The dataset includes detailed generation configurations such as base model, LoRA, sampler, steps, shift, CFG scale, frames, and seeds. The layout and file naming conventions are also described in the README.




