FocusFlow: Boosting Key-Points Optical Flow Estimation for Autonomous Driving

GitHub Link

The GitHub link is https://github.com/zhonghuayi/focusflow_official

Introduce

The GitHub repository “FocusFlow_official” presents FocusFlow, a framework designed to enhance key-point optical flow estimation for autonomous driving. The traditional data-driven optical flow methods often underperform in crucial key-point scenarios due to biased attention distribution. FocusFlow introduces a points-based modeling approach that explicitly learns key-point priors. The framework includes a mixed loss function combining photometric loss and Conditional Point Control Loss (CPCL) for diverse point-wise supervision. It employs a Condition Control Encoder (CCE) to replace the standard feature encoder, achieving remarkable precision improvement on various key points while maintaining competitive performance on entire frames. Contact for collaboration [email protected].

Based on the modeling method, we present FocusFlow, a framework consisting of 1) a mix loss function combined with a classic photometric loss function and our proposed Conditional Point Control Loss (CPCL) function for diverse point-wise supervision; 2) a conditioned controlling model which substitutes the conventional feature encoder by our proposed Condition Control Encoder (CCE).

Content

FocusFlow: Boosting Key-Points Optical Flow Estimation for Autonomous Driving __ We will release code and checkpoints in the future. __ Feel free to contact me if you have additional questions or have interests in collaboration. Please drop me an email at [email protected]. =)