task_categories:
- image-to-video
EPiC_Data: Training Data for Efficient Video Camera Control Learning
This repository contains the training data for EPiC: Efficient Video Camera Control Learning with Precise Anchor-Video Guidance. EPiC is an efficient and precise camera control learning framework that automatically constructs high-quality anchor videos without expensive camera trajectory annotations. This dataset provides the approximately 5K training videos and related components used to train the EPiC framework for image-to-video (I2V) and video-to-video (V2V) camera control tasks.
Paper: EPiC: Efficient Video Camera Control Learning with Precise Anchor-Video Guidance Project Page: https://zunwang1.github.io/Epic Code: https://github.com/wz0919/EPiC
Paper Abstract
Recent approaches on 3D camera control in video diffusion models (VDMs) often create anchor videos to guide diffusion models as a structured prior by rendering from estimated point clouds following annotated camera trajectories. However, errors inherent in point cloud estimation often lead to inaccurate anchor videos. Moreover, the requirement for extensive camera trajectory annotations further increases resource demands. To address these limitations, we introduce EPiC, an efficient and precise camera control learning framework that automatically constructs high-quality anchor videos without expensive camera trajectory annotations. Concretely, we create highly precise anchor videos for training by masking source videos based on first-frame visibility. This approach ensures high alignment, eliminates the need for camera trajectory annotations, and thus can be readily applied to any in-the-wild video to generate image-to-video (I2V) training pairs. Furthermore, we introduce Anchor-ControlNet, a lightweight conditioning module that integrates anchor video guidance in visible regions to pretrained VDMs, with less than 1% of backbone model parameters. By combining the proposed anchor video data and ControlNet module, EPiC achieves efficient training with substantially fewer parameters, training steps, and less data, without requiring modifications to the diffusion model backbone typically needed to mitigate rendering misalignments. Although being trained on masking-based anchor videos, our method generalizes robustly to anchor videos made with point clouds during inference, enabling precise 3D-informed camera control. EPiC achieves SOTA performance on RealEstate10K and MiraData for I2V camera control task, demonstrating precise and robust camera control ability both quantitatively and qualitatively. Notably, EPiC also exhibits strong zero-shot generalization to video-to-video scenarios.
Sample Usage
To download the EPiC training dataset, which consists of approximately 5,000 training videos, navigate to the data/train directory within your EPiC code repository clone, and use the following wget and unzip commands:
# Assuming you are in the root directory of the EPiC code repository, navigate to data/train
cd data/train
# Download the main training videos
wget https://huggingface.co/datasets/ZunWang/EPiC_Data/resolve/main/train.zip
unzip train.zip
(Optional) You can also download the pre-extracted VAE latents, which can save several hours of preprocessing time:
wget https://huggingface.co/datasets/ZunWang/EPiC_Data/resolve/main/train_joint_latents.zip
unzip train_joint_latents.zip
After downloading and unzipping, your data/train folder, following preprocessing steps (as described in the EPiC GitHub repository), should have a structure similar to this:
data/
└── train/
├── caption_embs/
├── captions/
├── joint_latents/
├── masked_videos/
├── masks/
└── videos/
For detailed instructions on preprocessing this data (e.g., extracting caption embeddings or VAE latents) and training the EPiC model, please refer to the EPiC GitHub repository.