EffOWT: Transfer Visual Language Models to Open-World Tracking Efficiently and Effectively

Open-World Tracking (OWT) aims to track every object of any category, which requires the model to have strong generalization capabilities. Trackers can improve their generalization ability by leveraging Visual Language Models (VLMs). However, challenges arise with the fine-tuning strategies when VLMs are transferred to OWT: full fine-tuning results in excessive parameter and memory costs, while the zero-shot strategy leads to sub-optimal performance. To solve the problem, EffOWT is proposed for efficiently transferring VLMs to OWT. Specifically, we build a small and independent learnable side network outside the VLM backbone. By freezing the backbone and only executing backpropagation on the side network, the model's efficiency requirements can be met. In addition, EffOWT enhances the side network by proposing a hybrid structure of Transformer and CNN to improve the model's performance in the OWT field. Finally, we implement sparse interactions on the MLP, thus reducing parameter updates and memory costs significantly. Thanks to the proposed methods, EffOWT achieves an absolute gain of 5.5% on the tracking metric OWTA for unknown categories, while only updating 1.3% of the parameters compared to full fine-tuning, with a 36.4% memory saving. Other metrics also demonstrate obvious improvement.

Scanning phase imaging without accurate positioning system

Ptychography, a high-resolution phase imaging technique using precise in-plane translation information, has been widely applied in modern synchrotron radiation sources across the globe. A key requirement for successful ptychographic reconstruction is the precise knowledge of the scanning positions, which are typically obtained by a physical interferometric positioning system. Whereas high-throughput positioning poses a challenge in engineering, especially in nano or even smaller scale. In this work, we propose a novel scanning imaging framework that does not require any prior position information from the positioning system. Specifically, our scheme utilizes the wavefront modulation mechanism to reconstruct the object functions at each scan position and the shared illumination function, simultaneously. The scanning trajectory information is extracted by our subpixel image registration algorithm from the overlap region of reconstructed object functions. Then, a completed object function can be obtained by assembling each part of the reconstructed sample functions. High-quality imaging of biological sample and position recovery with sub-pixel accuracy are demonstrated in proof-of-concept experiment. Based on current results, we find it may have great potential applications in high-resolution and high throughput phase imaging.