𝗗𝗮𝘆-𝟭𝟱𝟬 Computer Vision Learning 𝗛𝗶𝘁-𝗗𝗲𝘁𝗲𝗰𝘁𝗼𝗿: Hierarchical Trinity Architecture Search for Object Detection by Huawei Follow me for similar post : 🇮🇳 Ashish Patel Interesting Facts : 🔸 This is a paper in CVPR 2020 with over 32 citations. 🔸 It Outperforms with the backbone FPN, NAS-FPN, DetNAS, etc. ------------------------------------------------------------------- 𝗔𝗺𝗮𝘇𝗶𝗻𝗴 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 : https://lnkd.in/epka5Kx code : https://lnkd.in/e7GAfKA ------------------------------------------------------------------- 𝗜𝗠𝗣𝗢𝗥𝗧𝗔𝗡𝗖𝗘 🔸 NAS has made major breakthroughs in image recognition tasks, and Hit-Detector applies NAS to more complex target detection tasks. 🔸Before Hit-Detector, researchers have tried to apply NAS to target detection tasks, but most of the current NAS for Object Detection only focuses on Backbone (eg: DetNAS) or feature fusion methods (eg: NAS- FPN), and other components of the detection network are still manually designed. 🔸The author believes that this NAS+ manual design method will limit the performance of the detection network. The author unexpectedly discovered the flaws of the previous NAS. #computervision #artificialintelligence #innovation
Sign in to view more content
Create your free account or sign in to continue your search
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
🔸Hit-Detector naturally solves this flaw. It can not only search for the backbone, neck and head of the detection network at the same time, but also know which operations backbone, neck and head like to use to compose itself. 🔸The experimental results of Hit-Detector are also amazing. The coco data set only needs 27M parameters and does not use any tricks (without bells and whistles) to get 41.4 mAP. For previous post visit this github : https://github.com/ashishpatel26/365-Days-Computer-Vision-Learning-Linkedin-Post