This repository contains the testing code for the paper "TransZero++: Cross Attribute-guided Transformer for Zero-Shot Learning" submitted to TPAMI. We will release all codes of this work later.
We provide trained models (Google Drive) on three different datasets: CUB, SUN, AWA2 in the CZSL/GZSL setting. You can download model files as well as corresponding datasets, and organize them as follows:
.
├── saved_model
│ ├── TransZeroPP_CUB_CZSL.pth
│ ├── TransZeroPP_CUB_GZSL.pth
│ ├── TransZeroPP_SUN_CZSL.pth
│ ├── TransZeroPP_SUN_GZSL.pth
│ ├── TransZeroPP_AWA2_CZSL.pth
│ └── TransZeroPP_AWA2_GZSL.pth
├── data
│ ├── CUB/
│ ├── SUN/
│ └── AWA2/
└── ···
The code implementation of TransZero++ mainly based on PyTorch. All of our experiments run and test in Python 3.8.8. To install all required dependencies:
$ pip install -r requirements.txt
Runing following commands and testing TransZero++ on different dataset:
CUB Dataset:
$ python test.py --config config/CUB_CZSL.json # CZSL Setting
$ python test.py --config config/CUB_GZSL.json # GZSL Setting
SUN Dataset:
$ python test.py --config config/SUN_CZSL.json # CZSL Setting
$ python test.py --config config/SUN_GZSL.json # GZSL Setting
AWA2 Dataset:
$ python test.py --config config/AWA2_CZSL.json # CZSL Setting
$ python test.py --config config/AWA2_GZSL.json # GZSL Setting
Results of our released models using various evaluation protocols on three datasets, both in the conventional ZSL (CZSL) and generalized ZSL (GZSL) settings.
| Dataset | Acc(CZSL) | U(GZSL) | S(GZSL) | H(GZSL) |
|---|---|---|---|---|
| CUB | 78.3 | 67.5 | 73.6 | 70.4 |
| SUN | 67.6 | 48.6 | 37.8 | 42.5 |
| AWA2 | 72.6 | 64.6 | 82.7 | 72.5 |
Note: All of above results are run on a server with an AMD Ryzen 7 5800X CPU and a NVIDIA RTX A6000 GPU.
Parts of our codes based on:
If you have any questions about codes, please don't hesitate to contact us by gchenshiming@gmail.com or hoongzm@gmail.com.