Hello everyone , I meet you again , I'm your friend, Quan Jun .
pig_face_recognition
JD.COM JDD Pig face recognition competition
pytorch-baseline
1. Running environment
TeslaK20c Cluster single node dual card Red Hat 4.4.7-3 Python 2.7.13 cuda 8.0 cudnn 5.0 pytorch 0.3.0
2. Take the pig from the video
use yolo-9000 Algorithm , Manual beating label after , Yes ffmpeg The extracted video frame is used for pig target detection , Frame the main part of the pig , Make a foundation for subsequent classification .
3. Data preprocessing
Preprocess the downloaded data set , Generate torchvision.datasets.ImageFolder The folder format required by the interface , And use torchvision.transforms Data enhancement using the method in , Specifically used RandomResizedCrop、RandomHorizontalFlip、ColorJitter as well as Normalize.
4.train from scratch or fitune from imagenet
Try a variety of resnet and densenet The Internet , Finally, better results are selected for average fusion .
5. test result
Carry out forward operation on the test data set to obtain the predicted classification probability and save it .
6. Submit results
Convert score format to desired format and submit results .
Publisher : Full stack programmer stack length , Reprint please indicate the source :https://javaforall.cn/151872.html Link to the original text :https://javaforall.cn