AdaptIS: Adaptive Instance Selection Networkhtml
2019-09-19 12:58:07git
Paper: https://arxiv.org/pdf/1909.07829.pdf github
Code (MXNet): https://github.com/saic-vul/adaptis 函數
Pretrained model for ToyV1: https://drive.google.com/open?id=1IuJUh0JvbKYILBxCeO2h6U4LG-9DoTHi
Pretrained model for ToyV2: https://drive.google.com/open?id=1RxepfpJF5gRpRNYu1urdV748suF3TL5k優化
Related Paper: ui
Panoptic Segmentation, Alexander Kirillov, Kaiming He, Ross Girshick, Carsten Rother, Piotr Dollar google
Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization [Paper] [Code] [Blog] spa
An intriguing failing of convolutional neural networks and the CoordConv solution [Paper] [Code] [Blog] orm
1. Background and Motivation: htm
本文提出一種新的分割方式,即:給出一個 BBox,該方法能夠將該位置的物體分割出來,而不是所有分割出來。示意圖以下所示:
本文所提出方法的名稱爲:AdaptIS,不依賴於 bounding box proposal。而是直接優化目標分割精度。給定一張圖像 I 和 一個固定的 point proposal (x, y),做者直接優化目標損失函數。咱們利用一個 pixel-wise loss 來計算 AdaptIS 預測 和 target object 的 mask。