𝗗𝗮𝘆-𝟭𝟰𝟵 Computer Vision Learning 𝗔𝘂𝗴𝗙𝗣𝗡: Improving Multi-scale Feature Learning for Object Detection by Horizon Robotics and CASIA Follow me for similar post : 🇮🇳 Ashish Patel Interesting Facts : 🔸 This is a paper in CVPR 2020 with over 37 citations. 🔸 It Outperforms with the backbone Darknet-19, Resnet-101,Resnet-101-FPN, Inception-Resnet-V2,MobileNet-v2-FPN, ResNext-101-32x4d-FPN, ResNext-101-64x4d-FPN etc. ------------------------------------------------------------------- 𝗔𝗺𝗮𝘇𝗶𝗻𝗴 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 : https://lnkd.in/e8dbff5 code : https://lnkd.in/eVXdJGm ------------------------------------------------------------------- 𝗜𝗠𝗣𝗢𝗥𝗧𝗔𝗡𝗖𝗘 🔸 Current state-of-the-art detectors typically exploit feature pyramids to detect objects at different scales. Among them, FPN is one of the representative works that build a feature pyramid by multi-scale features summation. 🔸 AugFPN consists of three components: Consistent Supervision, Residual Feature Augmentation, and Soft RoI Selection. 🔸 Three improvement points are proposed for defects, which are exactly the three components of AugFPN: 1️⃣ Consistent Supervision: Used to reduce the semantic gap between different scales 2️⃣ Residual Feature Augmentation: used to reduce information loss in fusion, summation of different scales 3️⃣ Soft RoI Selection: better extract ROI Feature from the image pyramid for classification #computervision #artificialintelligence #innovation
🔸 Replacing FPN with AugFPN in Faster R-CNN, our models achieve 2.3 and 1.6 points higher Average Precision (AP) when using ResNet50 and MobileNet-v2 as backbone respectively. 🔸 AugFPN improves RetinaNet by 1.6 points AP and FCOS by 0.9 points AP when using ResNet50 as the backbone. For previous post visit this Github : https://github.com/ashishpatel26/365-Days-Computer-Vision-Learning-Linkedin-Post
University of Electronic…•1K followers
4yI really enjoy your posts, thanks a lot.