數據集二分類 第一類1000張, 第二類600張git
1. darknet + resnet50 Loss,訓練出來測試的時候是NULL, 暫時不知道爲何, 將CUDA-10.0 換成 cuda-8.0 依然顯示不出label Loss 也是3.8github
2. darknet + alexnet Loss = 3.8 , 測試
3. keras + inceptionV3 3d
1.) OpenCV4.0 調用失敗, 提示FusedBatchNorm is_learning = Truecode
2) https://github.com/opencv/opencv/issues/14236 修改代碼,修復bug後,出現下面bug DataType = DT_BOOL 不被識別orm
3) 須要更新到最新版的opencvblog
4. keras + inception_resnetV2 同上 get
6. tensorflow + slim + inceptionV4 驗證集準確率90%以上 , 可是OpenCV4.0 調用失敗, slim 竟然把DecodeJpeg層都加到 pb模型裏去了, 暫時不知道怎麼修改模型input
1) 修改成it
opencv 報錯 :
2)修改成
opencv ---in 'Mul'
3) 修改成
opencv --- 錯誤同上
4) modify as
call in opencv , error shows as above
5)
sh 腳本提示不能使用Cast節點做爲輸入 , 必須用Placeholder , 因而下面更改input
6)
opencv Same as above
7)
Error
8)
fail in ‘Mul’ as well
9)
fail in ‘Mul’ as well
10)
fail in ‘Mul’ as well
7. 準確率比較
keras + mobilenetv2 + 224X224 val_acc = 91% epoch=20
keras + mobilenetv2 + 448X448 val_acc = 97% epoch=40
keras + mobilenetv2 + 640X360 val_acc = 79% epoch=45 and val_acc = 74% epoch=20