Pytorch implement of "Dual-Space Contrastive Learning for Open-World Semi-Supervised Classification" in TNNLS 2024 (under review).
To start up, you need to install some packages. Our implementation is based on PyTorch. We recommend using conda to create the environment and install dependencies and all the requirements are listed in ./requirements.txt.
Preparation for Dataset and Pretrained Parameters
- The data preparation is the same as ORCA link. For cifar100 and cifar10, they will be downloaded automantically and you don't need to configure the dataset. For the Imagenet100, you need to download the ImageNet dataset first and then generate corresponding splitting lists via
gen_imagenet_list.py. - For We use SimCLR for pretraining. The weights used in our paper can be downloaded in this link.
- The pretrained checkpoints will be saved in
./pretrained/while the datasets will be saved in./datasets/. These settings are the same with the ORCA link. - You also need to create a empty folder name ```./checkpoints/''' under the root path to store the parameters of the model.
Get Started
-
For training on a certain dataset, you can config the argument
datasetas one ofcifar10,cifar100,imagenet100. Note that the other argument likelabeled-numandlabeled-ratioare fixed in the code. If you want to config it, please refer to the code. -
Our code automatically selects whether to run on a single GPU or multiple GPUs based on the number of devices specified in CUDA_VISIBLE_DEVICES.
-
For example, to train on CIFAR-100 with 50 labeled samples and 0.5 labeled ratio on GPU 0, you can run the following command:
export CUDA_VISIBLE_DEVICES=0
python train.py --dataset cifar100 --labeled-num 50 --labeled-ratio 0.5 --name 'cifar100_experiment'We use the MIT License. Details see the LICENSE file.
If you have any questions, you can turn to the issue block or send emails to 1120220290@mail.nankai.edu.cn. Glad to have a discussion with you!