zhanghx

关于训练数据

1

你好，感谢你的工程，在排序对比损失任务中，label.txt文件中每一行可以是不同退化类型的图片吗？比如 曝光 和 高斯噪声放一行，按照他们的mos分数做排序，感谢

Build Caffe and pycaffe

3

when build caffe and pycaffe: ``` voole@voole-OptiPlex-3020:~/fast-rcnn/caffe-fast-rcnn$ make -j8 && make pycaffe CXX .build_release/src/caffe/proto/caffe.pb.cc CXX src/caffe/blob.cpp CXX src/caffe/data_transformer.cpp CXX src/caffe/net.cpp CXX src/caffe/solver.cpp CXX src/caffe/util/insert_splits.cpp CXX src/caffe/util/db.cpp CXX src/caffe/util/upgrade_proto.cpp In file...

about the picture

1

hi first of all .thanks for your sharing .it is so helpful for us. but the picture cannot be seen.

Hi. thanks for your cool work. Could you please give me some information about the auto label?some key word or paper ? thanks very much.

input size:224 224 scale :True rotation:True 

do you use the place365 model to finetune your train?

关于训练

7

你好 我的数据分4类，每类10w左右，我finetune 所有layer，batchsize 128，lr 0.05 我发现迭代2个epoch模型的val acc 就可以到0.98左右，没感觉到是过拟合，但是实际测试的时候效果不是很好，许多正常的图片也会被误识别为性感的图片而且置信度还很高，感觉很怪，难道是网络过拟合了？但是观察val上的表现又不像，不知道您有没有遇到过这种情况，感谢

d_loss decline and g_loss rise

2

hello. First tanks for your code. When i train the model for about 10k iterations 6 epoch.the d_loss decline and g_loss rise.as the pic show:  and this is my...

about the shift module inplace output

3

I find that the output of inplace shift is different from the normal shift. is that right?

what is the setting of the param :DOUBLE_BIAS and WEIGHT_DECAY

4

First of all. thanks for your code. I don't know the setting about hte param : DOUBLE_BIAS and WEIGHT_DECAY . it show me that: **[epoch 0][iter 10] loss: nan RMSElog:...