jiewen

FER2013 resolution is 48*48, why input model resize to (224, 224), this operate will get blur or damage image quality?

so i use this method: - require_flatten=include_top it can run success

thanks, can you explain the rest question ?

thanks , that is fisrt train attribute classification layer ,get Att_fea, optimize the loss_att ; and then use the (b)net optimize the loss_reg and loss_rank?

> 暂时发现了一个问题，在rank_layer_live.py的get_minibatch(self,minibatch_db)函数里，minibatch_db是两幅图像的地址，jobs是得到的两幅3x224x224图像数据的列表，应该是读图失败了，读出来的是一个None，可能是它导致的错误。因为这一步用的是self.workers.map(preprocess,minibatch_db)，我看不到具体的处理过程，我想用其他的图像库读取，但是我不清楚你是读的整张图，然后resize成224x224x3，然后转成3x224x224的，还是读取整张图，随机截取224x224的范围，然后再转成3x224x244的。我不太清楚你是怎么得到jobs的，你能告诉我一下吗？ - 你好你的问题解决了吗？我训练的时候遇到相同的问题： ``` File "/home/xxx/desktop/RankIQA/src/data_layer/rank_layer_live.py", line 97, in forward blobs = self.get_minibatch(minibatch_db) File "/home/xxx/desktop/RankIQA/src/data_layer/rank_layer_live.py", line 72, in get_minibatch jobs =self.workers.map(preprocess,minibatch_db) File "/usr/lib/python2.7/multiprocessing/pool.py", line 251, in map return self.map_async(func,...

and can you release the pre train mode ?thanks!

@AakashKumarNain can you share how to run the evaluate.py, i don't understand the parser operator, how to add test image?

in this project, in squid directory only `train.py` used. `run.py` and `test.py` is not for this project, so you can ignore this error.

@Vipermdl this is my: ``` 2018-12-19 05:29:12,675 - root - INFO - Epoch:80, train PA1:0.960404015835341, MPA1:0.8597161731054243, MIoU1:0.7849944881184294, FWIoU1:0.926682433172048 2018-12-19 05:29:12,686 - root - INFO - validating one epoch... 2018-12-19 05:29:44,044...

@vineetgarg93 @EchoAmor i have train the model three months ago. i check the train log as follow: 点击查看详细内容 ``` 2018-12-18 17:54:05,198 - root - INFO - Global configuration as follows:...