Python影像辨識筆記(十)：使用GluonCV快速執行SSD、Faster RCNN、YOLOv3推論
介紹
GluonCV是一個專門用來執行電腦視覺任務的Python套件，功能相當廣泛。在Image Classification, Object Detection, Instance Segmentation, Semantic Segmentation, Pose Estimation, Action Recognition, Object Tracking, Dataset reparation, Distributed Training, Deployment都有不錯的應用。
我也有看到該模組能與OpenCV結合，達成即時Object Dection的功能。
本文將著重在如何透過Pre-train Model推論的部份，Model Training在官方的Tutorial也有詳細的介紹。
安裝gluoncv
# Cuda 10.1pip install --upgrade mxnet-cu101 gluoncv註：如果Cuda版本是10.0，則改成mxnet-cu100conda uninstall hdf5
conda install hdf5
官方教學
GluonCV討論版
Object Dection:
Predict with pre-trained YOLO models
API Difference between SSD, RCNN, YOLOx, img = data.transforms.presets.ssd.load_test(im_fname, short=512)
x, orig_img = data.transforms.presets.rcnn.load_test(im_fname)
x, img = data.transforms.presets.yolo.load_test(im_fname, short=512
04. Train SSD on Pascal VOC dataset
05. Deep dive into SSD training: 3 tips to boost performance
06. Train Faster-RCNN end-to-end on PASCAL VOC
07. Train YOLOv3 on PASCAL VOC
08. Finetune a pretrained detection model
09. Run an object detection model on your webcam
10. Skip Finetuning by reusing part of pre-trained model
11. Predict with pre-trained CenterNet models
For more usage, please refer to the official tutorial