filter_roidb(roidb) not found in train.py

[saraghaie]

Hello,

thank you for sharing your code. The function filter_roidb() is not available or imported in the lib/tdcnn/train.py but called at line 117 of train.py could you please provide this function or shall I just ignore it by commenting this line?

[huijuan88]

comment this line.

On Apr 5, 2018, at 11:03, saraghaie notifications@github.com<mailto:notifications@github.com> wrote:

Hello,

thank you for sharing your code. The function filter_roidb() is not available or imported in the lib/tdcnn/train.py but called at line 117 of train.py could you please provide this function or shall I just ignore it by commenting this line?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/VisionLearningGroup/R-C3D/issues/19, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AFOa_1BaPZQcPIiVhyNJsy4pAygeXp_rks5tljJIgaJpZM4TInrO.

[saraghaie]

Thank you for your quick answer!

[fanw52]

hello,I also come across this question,have you solve it? @saraghaie

[saraghaie]

@kstys yes, you can just comment the line and it will work :)

[fanw52]

thanks a lot @saraghaie

[fanw52]

and then a come across another question,

import google.protobuf as pb2 pb2.text_format.Merge(f.read(), self.solver_param)

AttributeError: 'module' object has no attribute 'text_format'

and I look for some solution,but fail,have you some suggestion?? @saraghaie

[fanw52]

thank you , @saraghaie and I have solve it, i add

import google.protobuf.text_format before

import google.protobuf as pb2 it will work

[saraghaie]

Hi @kstys, I didn't have any problem with import google.protobuf.text_format. Its perfect that you found the solution, it seems that some people also had the same issue after trying to use faster RCNN: https://github.com/rbgirshick/py-faster-rcnn/issues/198

[fanw52]

thanks a lot,have you train it on THUMOS14?? @saraghaie

[saraghaie]

@kstys yes I trained on THUMOS14 and I could get approximately the same results as the authors. But I needed to change some parts of the code for myself, because it is at some parts hard coded. Feel free to let me know if there is anything that I can help with.

[jxiangli]

Hi, @saraghaie The results are lower than the authors using the model trained on THUMOS14 by myself. Did you change the training parameters? And could you please tell me what parts of code you changed in detail?

[fanw52]

That's good ,and I want to train it on THUMOS14,but I can not find the file Class_Index_Detection.txt ,could you give me some hint and let me start the work ? @saraghaie

[saraghaie]

@kstys this file is provided with the THUMOS14 dataset, following is the content of Class_Index_Detection.txt

7 BaseballPitch 9 BasketballDunk 12 Billiards 21 CleanAndJerk 22 CliffDiving 23 CricketBowling 24 CricketShot 26 Diving 31 FrisbeeCatch 33 GolfSwing 36 HammerThrow 40 HighJump 45 JavelinThrow 51 LongJump 68 PoleVault 79 Shotput 85 SoccerPenalty 92 TennisSwing 93 ThrowDiscus 97 VolleyballSpiking

[saraghaie]

@jxiangli following are my settings in the td_cnn_end2end.yml. The changes that I applied were just for making the code run on my system, I don't know if such settings are necessary for everyone.

TRAIN: HAS_RPN: True VIDEO_BATCH: 1 TWIN_NORMALIZE_TARGETS_PRECOMPUTED: True RPN_POSITIVE_OVERLAP: 0.7 RPN_BATCHSIZE: 64 PROPOSAL_METHOD: gt BG_THRESH_LO: 0.0 RPN_MIN_SIZE: 8 LENGTH: [512] TEST: HAS_RPN: True NMS: 0.3 RPN_MIN_SIZE: 8 LENGTH: [512]

INPUT: "video" NUM_CLASSES: 21

[fanw52]

Thanks, I have sovle the all bug,and network has run,next,I will try to change some code @saraghaie

[jxiangli]

Thanks a lot!

[fanw52]

Hi, @saraghaie I have a problem about dataset that I can not understand. when I use the dataset THUMOS14 and then split into frames. and in the test data ,it only includes 213 files,but I have 1574 videos ,and then I look for TH14_Temporal_Annotations_Test ,and I find that it maybe only include 213 annotation videos ,but what 's the meaning of the remaining of 1361 videos ?

[ranranxu95]

how can i get this code?

[huijuan88]

The remaining videos may be used for other classification tasks.

It is in the preprocessing code.

On Jul 3, 2018, at 07:14, ranran notifications@github.com<mailto:notifications@github.com> wrote:

how can i get this code?

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/VisionLearningGroup/R-C3D/issues/19#issuecomment-402115186, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AFOa_9ZN9OYQN6Zf7_mLk5jBnhcFgUQOks5uC1IbgaJpZM4TInrO.