`

Timezone: »

 
Poster
Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data
Ashraful Islam · Chun-Fu (Richard) Chen · Rameswar Panda · Leonid Karlinsky · Rogerio Feris · Richard J. Radke

Thu Dec 09 04:30 PM -- 06:00 PM (PST) @ None #None

Most existing works in few-shot learning rely on meta-learning the network on a large base dataset which is typically from the same domain as the target dataset. We tackle the problem of cross-domain few-shot learning where there is a large shift between the base and target domain. The problem of cross-domain few-shot recognition with unlabeled target data is largely unaddressed in the literature. STARTUP was the first method that tackles this problem using self-training. However, it uses a fixed teacher pretrained on a labeled base dataset to create soft labels for the unlabeled target samples. As the base dataset and unlabeled dataset are from different domains, projecting the target images in the class-domain of the base dataset with a fixed pretrained model might be sub-optimal. We propose a simple dynamic distillation-based approach to facilitate unlabeled images from the novel/base dataset. We impose consistency regularization by calculating predictions from the weakly-augmented versions of the unlabeled images from a teacher network and matching it with the strongly augmented versions of the same images from a student network. The parameters of the teacher network are updated as exponential moving average of the parameters of the student network. We show that the proposed network learns representation that can be easily adapted to the target domain even though it has not been trained with target-specific classes during the pretraining phase. Our model outperforms the current state-of-the art method by 4.4% for 1-shot and 3.6% for 5-shot classification in the BSCD-FSL benchmark, and also shows competitive performance on traditional in-domain few-shot learning task.

Author Information

Ashraful Islam (Rensselaer Polytechnic Institute)

I am a final year PhD student at the ECSE department of Rensselaer Polytechnic Institute (RPI), Troy, NY, supervised by Prof. Richard Radke. Broadly, I am interested in computer vision and deep learning. I develop neural network models for unsupervised, semi-supervised, weakly-supervised, and few-shot learning. I have also worked on action detection and object tracking. Previously, I spent three wonderful summers at Microsoft Research, IBM Research and Kitware. I completed my undergrad in Electrical Engineering from BUET, Bangladesh.

Chun-Fu (Richard) Chen (MIT-IBM Watson AI Lab)
Rameswar Panda (MIT-IBM Watson AI Lab)
Leonid Karlinsky (Weizmann Institute of Science)
Rogerio Feris (MIT-IBM Watson AI Lab, IBM Research)
Richard J. Radke (Rensselaer Polytechnic Institute)

More from the Same Authors