Ashish Patel 🇮🇳’s Post

𝗗𝗮𝘆-𝟰𝟱𝟰 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗩𝗶𝘀𝗶𝗼𝗻 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 Balanced MSE for Imbalanced Visual Regression by  Follow me for a similar post: Ashish Patel ------------------------------------------------------------------- 𝗜𝗻𝘁𝗲𝗿𝗲𝘀𝘁𝗶𝗻𝗴 𝗙𝗮𝗰𝘁𝘀 : 🔸 This paper is published cvpr2022. 🔸 Github https://lnkd.in/gzyfcEAe ------------------------------------------------------------------- 𝗜𝗠𝗣𝗢𝗥𝗧𝗔𝗡𝗖𝗘 ➡️ Data imbalance exists ubiquitously in real-world visual regressions, e.g., age estimation and poses estimation, hurting the model's generalizability and fairness.  ➡️ Thus, imbalanced regression gains increasing research attention recently. Compared to imbalanced classification, imbalanced regression focuses on continuous labels, which can be boundless and high-dimensional and hence more challenging.  ➡️ In this work, we identify that the widely used Mean Square Error (MSE) loss function can be ineffective in imbalanced regression.  ➡️ We revisit MSE from a statistical view and propose a novel loss function, Balanced MSE, to accommodate the imbalanced training label distribution.  ➡️ We further design multiple implementations of Balanced MSE to tackle different real-world scenarios, particularly the one that requires no prior knowledge about the training label distribution.  ➡️ Moreover, to the best of our knowledge, Balanced MSE is the first general solution to high-dimensional imbalanced regression.  ➡️ Extensive experiments on both synthetic and three real-world benchmarks demonstrate the effectiveness of Balanced MSE.