This repository contains a PyTorch implementation of the mobile convolutional architecture given in ANTNets.
The following model comparison is with reference to ANTNet for ImageNet as given in the paper.
| Parameter | Official | Ours |
|---|---|---|
| Number of parameters (g = 1) | 3.7M | 3.81M |
| Number of parameters (g = 2) | 3.2M | 3.33M |
The cfg_antnet.py module has the model and training configuration in the form
of dict. This module can be passed to a training module to initialize the
model and kickstart training. An example module train.py is given for reference
and can be used as follows:
python3 train.py --cfg cfg_antnet.pyPlease feel free to write an issue in case you find a bug. Additional recommendations for features are welcome too!