Awesome

CREAM: Weakly Supervised Object Localization via Class RE-Activation Mapping

This is the official PyTorch implementation for CREAM. We provide the inference code for regressor-based CREAM on CUB dataset. The training code is coming soon. (Sorry our server was attacked and all the users were deleted. We are trying to get our latest code back.)

Requirements

pip install -r requirements.txt

Prepare Data

Download the official Caltech-UCSD-Birds-200-2011 (CUB) dataset.

Inference

mkdir checkpoints

Download the pretrained VGG16/InceptionV3 models for CUB and put them under checkpoints/:

• https://drive.google.com/drive/folders/1pgLjFN_LxKEA7dnmyVORUxFoEi_k9q86?usp=sharing

This example shows the usage of inference with vgg16 classification and localization model.

bash infer.sh

This inference code is based on PSOL (https://github.com/tzzcl/PSOL).