Bowtie Networks: Generative modeling for joint few-shot recognition and novel-view synthesis
This is the repository for Bowtie Networks: Generative modeling for joint few-shot recognition and novel-view synthesis, published at ICLR 2021.
Discription
Due to the memory constrain, for the classification model, we first pretrain a feature extration network student network to transfer the images to feature vectors. The student network works on the 64 x 64
resolution and is trained with knowledge distillation teacher network on full-scale images.
After that, we store the feature vectors, together with downsampled images, to a numpy file. For the classification task, we train classifiers on the feature level.
How to run
The model run on two stages, train (gan.train) on base classes and few-shot tune (gan.few) on novel classes. See train_cars.sh for a sample training.
dataset and pre-trained student network model folder: here
Acknowledgement
Our code is based on the awesome work of Hologan. The parameters are almost the same as them.
Citation
@inproceedings{bao2021bowtie,
Author = {Zhipeng Bao, Yu-Xiong Wang and Martial Hebert},
Title = {Bowtie Networks: Generative modeling for joint few-shot recognition and novel-view synthesis},
Booktitle = {ICLR},
Year = {2021},
}