xiaohaoChen / rrc_detection

Accurate Single Stage Detector Using Recurrent Rolling Convolution
Other
361 stars 136 forks source link

training.txt contains images in testing.txt #14

Open leighton613 opened 7 years ago

leighton613 commented 7 years ago

Hi! Thanks for your code and paper.

Wonder why the testing.txt and training.txt are different from the description in the paper which says validation has 2741 images.

In the repo, training.txt contains all the car images (6684), and the testing.txt has a subset (503). How did you pick this subset? Should I remove them from the training set to get validation results?

Thanks ahead!

xiaohaoChen commented 7 years ago

Hi! The testing.txt was just set for running the script. Since the training.txt contains all the car images, we didn't test in the validation set while training. The testing.txt is invalid here. If you need to do some experiments with validation set, I can sent you the testing.txt and training.txt as we descripted in the paper.

leighton613 commented 7 years ago

Hi Xiaohao, Thanks so much for sharing! Here's my email: leighton.zxw@gmail.com ;)

ChenLi2013 commented 6 years ago

Hi XiaohaoChen, Could you please send the testing.txt and training.txt of car ,pedestrian,cyclist to me? Here's my email: lilichen0906@163.com ;

foralliance commented 6 years ago

@xiaohaoChen

'test_interval': 1000000,

Because of this setting, there is no test in the training process. That's what you said earlier."we didn't test in the validation set while training"???

foralliance commented 6 years ago

@leighton613 @xiaohaoChen @ChenLi2013 Hi Could you please send the testing.txt and training.txt of car ,pedestrian,cyclist to me? Here's my email: nedved-f@163.com many many thanks

hulaifeng commented 6 years ago

@xiaohaoChen @ChenLi2013 @leighton613 @foralliance Could you please send the testing.txt and training.txt of car ,pedestrian,cyclist to me? Here's my email: 201221100501@std.uestc.edu.cn thanks