Day-25 Computer Vision Learning ParseNet: Looking Wider to see better(Semantic Segmentation) Follow me for similar post : 🇮🇳 Ashish Patel Interesting Facts : 🔸 This is a paper in 2016 ICLR paper with more than 728 citations 🔸 ParseNet detect the global object context with Fully Convolution Network ------------------------------------------------------------------- 𝗔𝗺𝗮𝘇𝗶𝗻𝗴 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 : https://lnkd.in/ensn6q2 Caffe: https://bit.ly/3sNRozf keras : https://bit.ly/39eFHcX Tensorflow : https://bit.ly/3iKzit3 Pytorch : https://bit.ly/2MhglCi ------------------------------------------------------------------- 𝗜𝗠𝗣𝗢𝗥𝗧𝗔𝗡𝗖𝗘 🔸 Global Context uses a sliding noise to interfere with the input image and observe the output of the network to detect how large the effective receptive field of a network is. 🔸 This is a good idea, because most of the papers use nuclear and other parameters to infer the receptive field, but how big is the really effective receptive field? It was found that by authors, in theory, should have fc7 VGG of the receptive field, but in fact only an image . 🔸 The author found that using a GlobalPooling can significantly increase the receptive field and improve the segmentation effect. #artificialintelligence #computervision
For All previous post visit: https://github.com/ashishpatel26/365-Days-Computer-Vision-Learning-Linkedin-Post
🔸 Other two Important things are EARLY FUSION AND LATE FUSION and L2 Norm Layer
I really appreciate your high tech posts Ashish!