Python code for detecting and learning new classes of threats present in crops

A toolbox for one-class classification and open set recognition based on intra-class splitting

Official PyTorch implementation of the paper “Catching Both Gray and Black Swans: Open-set Supervised Anomaly Detection”, open-set anomaly detection, few-shot anomaly detection, semi-supervised anomaly detection.

Machine learning project conducted together with Volvo Cars

[ECCV'22 Oral] Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentation on Complex Urban Driving Scenes. Dealing with out-of-distribution detection or open-set recognition in semantic segmentation.

Official implementation of "Extreme Value Meta-Learning for Few-Shot Open-Set Recognition of Hyperspectral Images" (TGRS'23)

Fooling Machine Learning Models: A Novel Out-of-Distribution Attack through Generative Adversarial Networks

Official PyTorch implementation for "SIO: Synthetic In-Distribution Data Benefits Out-of-Distribution Detection"

PyTorch implementation of our CVPR 2024 paper "Unified Entropy Optimization for Open-Set Test-Time Adaptation"

[IJCV 2022] Pytorch codes for Open-set Adversarial Defense with Clean-Adversarial Mutual Learning

1st Place Code for FungiCLEF 2023 Competition from UstcAIGroup

Source code for baseline obtenience

Official Implementation of "MORGAN: Meta-Learning-based Few-Shot Open-Set Recognition via Generative Adversarial Network" (WACV23)

Official Implementation of "Few-Shot Open-Set Recognition of Hyperspectral Images with Outlier Calibration Network" (WACV22)

Open set classification of car models. This 3-step classifier solves the problem where dogs are classified as cars, by first filtering these images out using ResNet CNNs transfer-trained on different datasets.

Open-Set Support Vector Machines (OSSVM) [see commit message https://github.com/pedrormjunior/ossvm/commit/50d51dc482c8e13df7d9037976b97db7e60a1ccf for usage]

VLG: General Video Recognition with Web Textual Knowledge (https://arxiv.org/abs/2212.01638)

This is a novel unknown sar target identification method based on feature extraction networks and KLD-RPA joint discrimination. Experiment results form MSTAR dataset shows that our proposed Fea-DA achieves state of the art unknown sar target identification accuracy while maintaining the high recognition accuracy of known target.

Official implementation of KDD'23 paper "Deep Weakly-supervised Anomaly Detection"