EdwardVincentMa

Results 13 comments of EdwardVincentMa

Test set: Average loss: 18154214.1728, Accuracy: 1000/10000 (10.00%) Best Accuracy: 77.37% Train Epoch: 175 [0/50000 (0%)] Loss: 15677834.000000 LR: 0.0001 Train Epoch: 175 [3200/50000 (6%)] Loss: 4205774.000000 LR: 0.0001 Train...

> It is recommend to set resize width and height with 0 (keep original size) when produce lmdb . Yes, I used the src size, but failed. ![image](https://user-images.githubusercontent.com/39480728/69946831-50b26d80-1527-11ea-84b5-ef0a4b01f267.png)

Sorry, I used Win10, not Linux. I generate LMDB with convert_imageset.exe(.cpp) not the create_data.sh

> 你可以试着把coco数据集的标注框画在图片上,你再观察一下数据框怎么样,你或许会有一些眉目,我个人得到的结论是coco数据集有一些是标注错误的框 你好,画在图上了,是正确的,

@eric612 Eric, 你好,请问能否给我一个你的 yolov3_coco_train.prototxt,也就是你的第一步训练coco,mAP=0.4的做参照,不胜感激,因为我训练的mAP实在很低,不知道哪里出问题。

> [train.txt](https://github.com/eric612/MobileNet-YOLO/files/3976489/train.txt) > [solver.txt](https://github.com/eric612/MobileNet-YOLO/files/3976490/solver.txt) > [test.txt](https://github.com/eric612/MobileNet-YOLO/files/3976491/test.txt) > [train_moblienetv2_yolov3_coco.sh.txt](https://github.com/eric612/MobileNet-YOLO/files/3976500/train_moblienetv2_yolov3_coco.sh.txt) 请问您的coco数据集有经过筛选或者处理吗,coco本身有很多垃圾数据

> 何謂垃圾數據?且你的LOG看起來應該是因為pre-trained weight不對 你好,垃圾数据就是标签遗漏,或者标签错误,或者过小甚至根本看不清物体的也打上了标签。 我没用任何pre-trained weight,从头训练的。请问step1用了pretrained model了吗?

> Please use imagenet pre-trained weights > https://drive.google.com/file/d/0B3gersZ2cHIxZi13UWF0OXBsZzA/view > > Reference from > https://github.com/chuanqi305/MobileNet-SSD 谢谢,I will try, and then feedback. Thank you very much!

我用了pretrained caffemodel,也用了你给我的 train.txt,solver.txt,第一步训练 yolo_coco 的 mAP 还是20%左右。 train_net: "models/yolov3_coco/mobilenet_yolov3_coco_train.prototxt" test_net: "models/yolov3_coco/mobilenet_yolov3_coco_test.prototxt" test_iter: 5000 test_interval: 3000 base_lr: 0.0005 display: 10 max_iter: 160000 lr_policy: "multistep" gamma: 0.3 weight_decay: 0.000004 snapshot: 3000 snapshot_prefix:...