Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, ResNet |
Max Iter | 270000 |
SHOW MORE |
Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, ResNet |
ID | 164254221.0 |
SHOW MORE |
Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, ResNet |
ID | 164255101 |
SHOW MORE |
Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, ResNet |
ID | 164955410.0 |
SHOW MORE |
Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, ResNeXt |
Max Iter | 270000 |
SHOW MORE |
Training Techniques | SGD with Momentum, Random Horizontal Flip, Weight Decay |
---|---|
Architecture | PointRend, Mask R-CNN, FPN, SemanticFPN, ResNet |
ID | 202576688 |
SHOW MORE |
PointRend is a module for image segmentation tasks, such as instance and semantic segmentation, that attempts to treat segmentation as image rending problem to efficiently "render" high-quality label maps. It uses a subdivision strategy to adaptively select a non-uniform set of points at which to compute labels. PointRend can be incorporated into popular meta-architectures for both instance segmentation (e.g. Mask R-CNN) and semantic segmentation (e.g. FCN). Its subdivision strategy efficiently computes high-resolution segmentation maps using an order of magnitude fewer floating-point operations than direct, dense computation. Most importantly, Faster R-CNN was not designed for pixel-to-pixel alignment between network inputs and outputs. This is evident in how RoIPool, the de facto core operation for attending to instances, performs coarse spatial quantization for feature extraction. To fix the misalignment, Mask R-CNN utilises a simple, quantization-free layer, called RoIAlign, that faithfully preserves exact spatial locations.
This Colab Notebook tutorial contains examples of PointRend usage and visualizations of its point sampling stages.
To train a model with 8 GPUs run:
cd /path/to/detectron2/projects/PointRend
python train_net.py --config-file configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_coco.yaml --num-gpus 8
Model evaluation can be done similarly:
cd /path/to/detectron2/projects/PointRend
python train_net.py --config-file configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_coco.yaml --eval-only MODEL.WEIGHTS /path/to/model_checkpoint
@InProceedings{kirillov2019pointrend,
title={{PointRend}: Image Segmentation as Rendering},
author={Alexander Kirillov and Yuxin Wu and Kaiming He and Ross Girshick},
journal={ArXiv:1912.08193},
year={2019}
}
MODEL | MASK AP |
---|---|
PointRend (X101-FPN, 3×) | 41.1 |
PointRend (R101-FPN, 3×) | 40.1 |
PointRend (R50-FPN, 3×) | 38.3 |
PointRend (R50-FPN, 1×) | 36.2 |
MODEL | MIOU |
---|---|
SemanticFPN + PointRend (R101-FPN) | 78.9 |
MODEL | MASK AP |
---|---|
PointRend (R50-FPN, 1×, Cityscapes) | 35.9 |