Ashish Patel 🇮🇳’s Post

𝗗𝗮𝘆-𝟯𝟳𝟱 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗩𝗶𝘀𝗶𝗼𝗻 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗥𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝗚𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗲𝗱 𝗖𝗮𝘁𝗲𝗴𝗼𝗿𝘆 𝗗𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝗯𝘆 𝗩𝗚𝗚 𝗚𝗿𝗼𝘂𝗽 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 𝗼𝗳 𝗢𝘅𝗳𝗼𝗿𝗱 𝗮𝗻𝗱 𝗧𝗵𝗲 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 𝗼𝗳 𝗛𝗼𝗻𝗴 𝗞𝗼𝗻𝗴 Follow me for a similar post: Ashish Patel ------------------------------------------------------------------- 𝗜𝗻𝘁𝗲𝗿𝗲𝘀𝘁𝗶𝗻𝗴 𝗙𝗮𝗰𝘁𝘀 : 🔸 Paper: 𝗥𝗲𝗮𝗹𝗶𝘀𝘁𝗶𝗰 𝗚𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗲𝗱 𝗖𝗮𝘁𝗲𝗴𝗼𝗿𝘆 𝗗𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 🔸 This paper is published arxiv2022. 🔸 Proposed and investigated GCD, a more general, challenging and realistic variant of NCD. Given a dataset with partial labels for images and classes, the task is to find the assignment for all unlabelled images, while discovering any new classes.This developed baselines for the task by transferring state-of-the-art algorithms from NCD and further developed a simple but effective algorithm which substantially outperforms these baselines and proposed a method for identifying the number of categories in the unlabelled data, which is an understudied problem of practical significance.  ------------------------------------------------------------------- 𝗜𝗠𝗣𝗢𝗥𝗧𝗔𝗡𝗖𝗘 🔸 In this paper, we consider a highly general image recognition setting wherein, given a labelled and unlabelled set of images, the task is to categorize all images in the unlabelled set.  🔸 Here, the unlabelled images may come from labelled classes or from novel ones. Existing recognition methods are not able to deal with this setting, because they make several restrictive assumptions, such as the unlabelled instances only coming from known --- or unknown --- classes and the number of unknown classes being known a-priori.  🔸 We address the more unconstrained setting, naming it `Generalized Category Discovery', and challenge all these assumptions.  🔸 We first establish strong baselines by taking state-of-the-art algorithms from novel category discovery and adapting them for this task.  🔸 Next, we propose the use of vision transformers with contrastive representation learning for this open world setting.  🔸 We then introduce a simple yet effective semi-supervised $k$-means method to cluster the unlabelled data into seen and unseen classes automatically, substantially outperforming the baselines.  🔸 Finally, we also propose a new approach to estimate the number of classes in the unlabelled data.  🔸 We thoroughly evaluate our approach on public datasets for generic object classification including CIFAR10, CIFAR100 and ImageNet-100, and for fine-grained visual recognition including CUB, Stanford Cars and Herbarium19, benchmarking on this new setting to foster future research. #computervision #artificialintelligence #innovation