Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
AngelWings1997 11fc5a68f6 | 1 year ago | |
---|---|---|
.. | ||
eval/pytorch | 1 year ago | |
models | 1 year ago | |
scripts | 1 year ago | |
utils | 1 year ago | |
.gitignore | 1 year ago | |
README.md | 1 year ago | |
catboost_example.py | 1 year ago | |
eval_net.py | 1 year ago | |
eval_net_pred.py | 1 year ago | |
extract_features.py | 1 year ago | |
infer_meta_classifier.py | 1 year ago | |
lightgbm_example.py | 1 year ago | |
stats.md | 1 year ago | |
tags_analysis.py | 1 year ago | |
train_net.py | 1 year ago | |
train_net_single_label.py | 1 year ago |
This is our attempt to solve real-time multi-label classification challenge organized on AiChallenger by MeiTu company.
Simple baseline is composed of the following steps:
python scripts/video_to_frames.py --ann_file /mnt/ssd1/dataset/short_video_trainingset_annotations.txt.082902 --data_dir /mnt/ssd1/dataset/train/ --out_dir /mnt/ssd1/dataset/train_jpg
python scripts/video_to_frames.py --ann_file /mnt/ssd1/dataset/short_video_validationset_annotations.txt.0829 --data_dir /mnt/ssd1/dataset/val/ --out_dir /mnt/ssd1/dataset/val_jpg
CUDA_VISIBLE_DEVICES=0 screen python extract_features.py --ann_file /mnt/ssd1/dataset/new_short_video_validationset_annotations.txt.txt --data_dir /mnt/ssd1/dataset/val_jpg/ --out_dir /mnt/ssd1/dataset/val_features
CUDA_VISIBLE_DEVICES=0 screen python extract_features.py --ann_file /mnt/ssd1/dataset/new_short_video_trainingset_annotations.txt.txt --data_dir /mnt/ssd1/dataset/train_jpg/ --out_dir /mnt/ssd1/dataset/train_features
python lightgbm_example.py --train_dir /mnt/ssd1/dataset/train_features/ --val_dir /mnt/ssd1/dataset/val_features/
python train_net.py --train_ann_file /mnt/ssd1/dataset/new_short_video_trainingset_annotations.txt.txt --train_data_dir /mnt/ssd1/dataset/train_jpg/ --val_ann_fil
e /mnt/ssd1/dataset/new_short_video_validationset_annotations.txt.txt --val_data_dir /mnt/ssd1/dataset/val_jpg/ -a se_resnet50
The better way
screen ./scripts/unstable/train_baseline.sh 4 se_resnet50 0.0001 128
CUDA_VISIBLE_DEVICES=2 python eval_net.py -b 32 -a se_resnet50 -w /mnt/ssd1/easygold/challengerai-mlsv2018/logs/un_baseline_2018-10-10_16-57-05/checkpoint.pth.tar --ann_fi
le /mnt/ssd1/dataset/short_video_validationset_annotations.txt --data_dir /mnt/ssd1/dataset/val/ -t 1
Look for the frequence of tags to determine what tags often meet together. May be useful for multilabel analysis.
python tags_analysis.py --ann_file short_video_trainingnset_annotations.txt --out_dir mnt/ssd1/dataser/train_tags_analysis
Run meta classifier. It is based on majority voting. You can include uo to 5 single model predictions(lightgbm, catboost, log regression, neural network from pickles. Time for one prediction is printed for every model.
CUDA_VISIBLE_DEVICES=2 python infer_meta_classifier.py --train_dir /mnt/ssd1/dataset/multi_class/6_frames_train_se_resnet_merged/ --val_dir /mnt/ssd1/dataset/multi_class/6_frames_val_se_resnet_merged/ -multi_label -val_1_frame -net_pred -lgb -logreg -svm -catboost -voting
Files required for building docker are stored here.
Following packages are required to run the code:
No Description
Python Text Shell Dockerfile Markdown
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》