S-Prompts

Evaluation code for S-Prompts "S-Prompts Learning with Pre-trained Transformers: An Occam’s Razor for Domain Incremental Learning"

Enviroment setup

Create a conda env:

conda env create -f environments.yaml

Getting pretrained models

Pretrained Model for CDDB:

https://drive.google.com/file/d/1MXoBpVnGe_1aHQRLL9-VcdkDoTfeXOiM/view?usp=sharing

Pretrained Model for CORe50:

https://drive.google.com/file/d/1HD7auESA89zOxdDUN0hFeAVPvOPWUAgV/view?usp=sharing

Pretrained Model for DomainNet:

https://drive.google.com/file/d/1qDWMnxyVXNCRCmNls1lrB3k4UleON_w4/view?usp=sharing

Preparing data

Please refer to the following links to download and prepare data.

CDDB:
https://github.com/Coral79/CDDB
CORe50:
https://vlomonaco.github.io/core50/index.html#dataset
DomainNet:
http://ai.bu.edu/M3SDA/

After unzipping downloaded files, the file structure should be as shown below.

DeepFake_Data
├── biggan
│   ├── test
│   ├── train
│   └── val
├── gaugan
│   ├── test
│   ├── train
│   └── val
├── san
│   ├── test
│   ├── train
│   └── val
├── whichfaceisreal
│   ├── test
│   ├── train
│   └── val
├── wild
│   ├── test
│   ├── train
│   └── val
... ...

core50
└── core50_128x128
    ├── labels.pkl
    ├── LUP.pkl
    ├── paths.pkl
    ├── s1
    ├── s10
    ├── s11
    ├── s2
    ├── s3
    ├── s4
    ├── s5
    ├── s6
    ├── s7
    ├── s8
    └── s9

domainnet
├── clipart
│   ├── aircraft_carrier
│   ├── airplane
│   ├── alarm_clock
│   ├── ambulance
│   ├── angel
│   ├── animal_migration
│   ... ...
├── clipart_test.txt
├── clipart_train.txt
├── infograph
│   ├── aircraft_carrier
│   ├── airplane
│   ├── alarm_clock
│   ├── ambulance
│   ... ...
├── infograph_test.txt
├── infograph_train.txt
├── painting
│   ├── aircraft_carrier
│   ├── airplane
│   ├── alarm_clock
│   ├── ambulance
│   ├── angel
│   ... ...
... ...

Launching experiments

python eval.py --resume ./deepfake.pth --dataroot [YOUR PATH]/DeepFake_Data/ --datatype deepfake 
python eval.py --resume ./core50.pth --dataroot [YOUR PATH]/core50_128x128 --datatype core50 
python eval.py --resume ./domainnet.pth --dataroot [YOUR PATH]/domainnet --datatype domainnet