PackNet: https://arxiv.org/abs/1711.05769

Pretrained models are available here: https://uofi.box.com/s/zap2p03tnst9dfisad4u0sfupc0y1fxt
Datasets in PyTorch format are available here: https://uofi.box.com/s/ixncr3d85guosajywhf7yridszzg5zsq
The PyTorch-friendly Places365 dataset can be downloaded from http://places2.csail.mit.edu/download.html
Place models in checkpoints/ and unzipped datasets in data/

| | VGG-16 LwF | VGG-16 | VGG-16 BN | ResNet-50 | DenseNet-121 | |:-------------:|:------------:|:------------:|:------------:|:------------:|:------------:| | ImageNet | 36.58 (14.75)| 29.19 (9.90) | 27.10 (8.70) | 24.33 (7.17) | 25.51 (7.85) | | CUBS | 34.24 | 22.56 | 20.43 | 19.59 | 20.11 | | Stanford Cars | 22.07 | 17.09 | 14.92 | 14.03 | 16.18 | | Flowers | 12.15 | 11.07 | 8.59 | 8.12 | 9.07 |

Note that the numbers in the paper are averaged over multiple runs for each ordering of datasets. The pretrained models are for a specific dataset addition ordering: (c) CUBS Birds, (s) Stanford Cars, (f) Flowers.

These numbers were obtained by evaluating the models on a Titan X (Pascal).
Note that numbers on other GPUs might be slightly different (~0.1%) owing to cudnn algorithm selection.
https://discuss.pytorch.org/t/slightly-different-results-on-k-40-v-s-titan-x/10064

Requirements:

Python 2.7 or 3.xx
torch==0.2.0.post3
torchvision==0.1.9
torchnet (pip install git+https://github.com/pytorch/tnt.git@master)
tqdm (pip install tqdm)

Training:

Check out the scripts in src/scripts.
Run all code from the src/ directory, e.g. ./scripts/run_all.sh

Eval:

cd src  # Run everything from src/# Pruning-based models.python main.py --mode eval --dataset cubs_cropped 
  --loadname ../checkpoints/csf_0.75,0.75,-1_vgg16_0.5-nobias-nobn_1.pt# LwF models.python lwf.py --mode eval --dataset cubs_cropped 
  --loadname ../checkpoints/csf_lwf.pt