feature_statistics_path

[LRLLRL]

feature_statistics_path: "examples/cvpr2017/mit67/image_retrieval/feature_extraction/mit67_alexnet_color_conv11_5_feature_stats_domain.txt"
I don't understand this address.Could you explain?

[ZYYSny]

It stands for the bounds of the soft histogram bins. Before extracting features，you should run g LRLLRL <[email protected]>于2018年3月13日 周二下午6:12写道：

…

feature_statistics_path: "examples/cvpr2017/mit67/image_retrieval/feature_extraction/mit67_alexnet_color_conv11_5_feature_stats_domain.txt" I don't understand this address.Could you explain? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4>, or mute the thread <https://github.com/notifications/unsubscribe-auth/APtvSfLOv0tK2UPYg94jB9P4Og4gfyUWks5td5uKgaJpZM4SoZ70> .

[ZYYSny]

It stands for the bounds of the soft histogram bins. Before extracting features，you should run get_feature_stats.sh twice. The first time is to get the feature value range of each channel. The second time is to get the soft bins of histograms. 戈维峰 <[email protected]>于2018年3月14日 周三上午8:24写道：

…

It stands for the bounds of the soft histogram bins. Before extracting features，you should run g LRLLRL ***@***.***>于2018年3月13日 周二下午6:12写道： > feature_statistics_path: > "examples/cvpr2017/mit67/image_retrieval/feature_extraction/mit67_alexnet_color_conv11_5_feature_stats_domain.txt" > > I don't understand this address.Could you explain? > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub > <#4>, or mute > the thread > <https://github.com/notifications/unsubscribe-auth/APtvSfLOv0tK2UPYg94jB9P4Og4gfyUWks5td5uKgaJpZM4SoZ70> > . >

[daa233]

Hi, @ZYYSzj, when running get_feature_stats.sh the first time, how to designate the previous_statistics_path in image_retrieval/feature_stats/stats_deploy.prototxt? Do we have one txt file for that parameter before we run it for the first time?

I am confused because I get an error of Check failed: idx == total_channels_ (0 vs. 1376) when run the script for the first time. Thanks for any help.

[17860846552]

@daa233 I got the same problem as you, have you solved this problem?

[daa233]

@17860846552 No, I haven't... 😢 Not very clear about some details in xxx_deploy.prototxt.

[ZYYSny]

@LRLLRL @daa233 @17860846552 I have updated the ReadMe.md in the get_the_features_statistics branch. I am busy in other projects now. Sorry for the delay.

[daa233]

@ZYYSzj Thanks for updating.
I think I have gotten the main idea. When we run the code for the first time, we need to provide fake files like the format of "caltech_alexnet_color_feature_stats.txt". Otherwise, we will get the above error Check failed: idx == total_channels_ (0 vs. 1376) since we don't provide the fake files. And before running the code for the second time, we should rename "caltech_alexnet_color_feature_stats.txt" as "caltech_alexnet_color_feature_stats_previous.txt".

[ZYYSny]

@daa233 You are right. Since the codes is not well written. You need to provide fake files in the same format above. If you have 1376 feature maps, you need randomly generate 1376 lines. Then you can run the script again. You just need to run over you training set one epoch when you run the scripts.

[17860846552]

@ZYYSzj It really works! Thank you for taking time out of your busy schedule.