mmclassification训练
train
生成imagenet数据集
修改模型配置文件resnext50_32x4d.py修改输出类别
修改datasets imagenet.py CLASSES列表变量
1
| CLASSES = ['Prim', 'Lym', 'Mono', 'Plas', 'Red', 'Promy', 'Myelo', 'Late', 'Rods', 'Lobu', 'Eosl']
|
1
| python tools/train.py configs/resnet/resnet50_b32x8_imagenet.py --work-dir work_dirs
|
1
| python tools/train.py configs/swin_transformer/swin_small_224_b16x64_300e_imagenet.py --work-dir work_dirs_swin
|
test
1
| python tools/test.py configs/resnet/resnet50_b32x8_imagenet.py work_dirs/latest.pth --out result_test/resnet.pkl --metrics precision --out-items all
|
分析结果
1
| python tools/analysis_tools/analyze_results.py configs/swin_transformer/swin_small_224_b16x64_300e_imagenet.py result_test/resnet.pkl --out-dir result_test
|
训练swin_transformer
1
| python tools/train.py configs/swin_transformer/swin_small_224_b16x64_300e_imagenet.py --work-dir work_dir_swin --resume-from work_dirss/latest.pth
|
测试swin_transformer
1
| python tools/test.py configs/swin_transformer/swin_small_224_b16x64_300e_imagenet.py work_dirs_swin/epoch_122.pth --out result_test/swin.pkl --metrics precision --out-items all
|
分析结果
1
| python tools/analysis_tools/analyze_results.py configs/swin_transformer/swin_small_224_b16x64_300e_imagenet.py result_test/swin.pkl --out-dir result_test
|
训练在mmdetection上修改vit-transformer，尝试修改成为transformer-FG
1
| python tools/train.py configs/vision_transformer/vit-base-p16_ft-64xb64_in1k-384.py --work-dir work_dir/vit_fine_grained --resume-from work_dir/vit/latest.pth
|