mmskeleton
mmskeleton copied to clipboard
open-mmlab
Training on a different dataset
Dataset used
Cooking Activity
Modified the data accordingly. The accuracies are not that great. I have tried changing the parameters multiple times, using pre trained models as well. Still no improvements. Also I get the same sequence of labels as output every time.
label_sequence = output.sum(dim=2).argmax(dim=0)
Any idea where it might be going wrong?
Hi, @Zzoey, Thank you for your interests in our work.
I roughly skim the dataset you mentioned. It seems to be much easier than Kinetics-skeleton dataset. [1] used a naive K-means approach and achieve a mean average precision (AP) of 34.6% on this dataset. For your information, we also evaluate the results of K-means approaches and SVM in our Kinetics-skeleton dataset (with different features), but we only achieved 14.5% (AP) thereon. And our ST-GCN reached 30.8% (AP).
What about the accuracy you achieved? And how many joints do the pose tracks of this dataset have? For using Graph Convolution in our ST-GCN, we defined the joint graph manually. If the definition of the human joints is different between Cooking Activity and Kinetics-skeleton, you may not be able to use our pre-trained model.
[1] A Database for Fine Grained Activity Detection of Cooking Activities, M. Rohrbach, S. Amin, M. Andriluka and B. Schiele, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June, (2012)
@yysijie Thanks for writing back and for the information. I have obtained joints by using openpose (COCO - 18) like the one you used on Kinetics. My accuracy is around 9%. Although most of the output is only one label. I am not sure why.. It would be great if you can help! Thank you.
@Zzoey, can you share the config file and command you used? I think that will give me some hints.
@Zzoey hello,I also want to know how skeleton key points should be stored in train_data.npy,and how to write the tain_label.pkl?Can you share you Script ，my Email [email protected]，thanks。
@yysijie how to get this file??thank you
Did you download kinetics-skeleton? If you already downloaded, you need to run like this command to extract these files. python tools/kinetics_gendata.py
@JasOlean I am sorry to interrupted to you. can you give the url of openpose build file in ubuntu to me? i can't compile the openpose, it waste my lots of time.thank you.
How to get this file of my new datasets?thank you.
---Original--- From: "JasOlean"[email protected] Date: Fri, Jun 7, 2019 19:17 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
@yysijie how to get this file??thank you
Did you download kinetics-skeleton? If you already downloaded, you need to run like this command to extract these files. python tools/kinetics_gendata.py
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
Not kinetics-skeleton dataset.
---Original--- From: "JasOlean"[email protected] Date: Fri, Jun 7, 2019 19:17 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
@yysijie how to get this file??thank you
Did you download kinetics-skeleton? If you already downloaded, you need to run like this command to extract these files. python tools/kinetics_gendata.py
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
You can download that dataset via their link.
On Sun, Jun 9, 2019 at 15:57 wei-li [email protected] wrote:
Not kinetics-skeleton dataset.
---Original--- From: "JasOlean"[email protected] Date: Fri, Jun 7, 2019 19:17 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
@yysijie how to get this file??thank you
Did you download kinetics-skeleton? If you already downloaded, you need to run like this command to extract these files. python tools/kinetics_gendata.py
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GDGRBFN57LX4LEHMD3PZSSV3A5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXIEWZY#issuecomment-500190055, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GB26OOHUDMN3JLSFETPZSSV3ANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
@JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much.
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png [image: 图片] https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png [image: 图片] https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
Ok, thank you. very much
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:14 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png [image: 图片] https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png [image: 图片] https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
By the way, do you know about Top 1 and Top 5 in their accuracy? Could you tell me about that what are they? I am not clear so much.
On Wed, Jun 12, 2019 at 13:18 wei-li [email protected] wrote:
Ok, thank you. very much
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:14 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] < https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png>
[image: 图片] < https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png>
[image: 图片] < https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png>
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub < https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707>,
or mute the thread < https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA>
.
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GDLQI3XOB76YHHH2ATP2B2JXA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPGMNI#issuecomment-501114421, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GFO3KQBWYOYZ3Z4OSDP2B2JXANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
You mean the code part?
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:21 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
By the way, do you know about Top 1 and Top 5 in their accuracy? Could you tell me about that what are they? I am not clear so much.
On Wed, Jun 12, 2019 at 13:18 wei-li [email protected] wrote:
Ok, thank you. very much
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:14 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] < https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png>
[image: 图片] < https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png>
[image: 图片] < https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png>
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub < https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707>,
or mute the thread < https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA>
.
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GDLQI3XOB76YHHH2ATP2B2JXA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPGMNI#issuecomment-501114421, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GFO3KQBWYOYZ3Z4OSDP2B2JXANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
No, I mean explanation about Top 1 and Top 5. What is the difference between them? I did not find in their paper about that.
On Wed, Jun 12, 2019 at 13:31 wei-li [email protected] wrote:
You mean the code part?
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:21 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
By the way, do you know about Top 1 and Top 5 in their accuracy? Could you tell me about that what are they? I am not clear so much.
On Wed, Jun 12, 2019 at 13:18 wei-li [email protected] wrote:
Ok, thank you. very much
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:14 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] <
https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png>
[image: 图片] <
https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png>
[image: 图片] <
https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png>
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <
https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707>,
or mute the thread <
https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA>
.
--
ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub < https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GDLQI3XOB76YHHH2ATP2B2JXA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPGMNI#issuecomment-501114421>,
or mute the thread < https://github.com/notifications/unsubscribe-auth/AEOY3GFO3KQBWYOYZ3Z4OSDP2B2JXANCNFSM4FPOGCFA>
.
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GFERBIP2HPQL6E25YTP2B3Y3A5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPG24Y#issuecomment-501116275, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GCAO7F32NPNGE5LWP3P2B3Y3ANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
I think it's just like the correct predict is appear among the top five predictions --top 5. it is also understand to top1
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:35 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
No, I mean explanation about Top 1 and Top 5. What is the difference between them? I did not find in their paper about that.
On Wed, Jun 12, 2019 at 13:31 wei-li [email protected] wrote:
You mean the code part?
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:21 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
By the way, do you know about Top 1 and Top 5 in their accuracy? Could you tell me about that what are they? I am not clear so much.
On Wed, Jun 12, 2019 at 13:18 wei-li [email protected] wrote:
Ok, thank you. very much
---Original--- From: "JasOlean"[email protected] Date: Wed, Jun 12, 2019 12:14 PM To: "yysijie/st-gcn"[email protected]; Cc: "Comment"[email protected];"wei-li"[email protected]; Subject: Re: [yysijie/st-gcn] Training on a different dataset (#81)
Do you have your own dataset like kinetics skeletons dataset? If you do not have, you have to create your Jason dataset. If you already have, you can just use their kinetic_gendata.py.
On Wed, Jun 12, 2019 at 11:20 wei-li [email protected] wrote:
@JasOlean https://github.com/JasOlean thank you reply on me. But i want to the Script to get .npy and .pkl file of our own datasets, just like this picture below(picture 1). i couldn't open this file by pycharm, just want to see this file contents(picture 2 and picture 3). Can you share the Script to me? thank you very much. [image: 图片] <
https://user-images.githubusercontent.com/32689594/59318845-134cf000-8cfb-11e9-9afe-250d9bd2bcf9.png>
[image: 图片] <
https://user-images.githubusercontent.com/32689594/59318940-776fb400-8cfb-11e9-8ce8-4bae3b3fc3a1.png>
[image: 图片] <
https://user-images.githubusercontent.com/32689594/59318951-822a4900-8cfb-11e9-8d7a-71d1a1ad5b32.png>
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <
https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GHFJEHX3WAAMFNEOK3P2BMOHA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPBSMY#issuecomment-501094707>,
or mute the thread <
https://github.com/notifications/unsubscribe-auth/AEOY3GGCA5VQRUITRBRKY7TP2BMOHANCNFSM4FPOGCFA>
.
--
ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub < https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GDLQI3XOB76YHHH2ATP2B2JXA5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPGMNI#issuecomment-501114421>,
or mute the thread < https://github.com/notifications/unsubscribe-auth/AEOY3GFO3KQBWYOYZ3Z4OSDP2B2JXANCNFSM4FPOGCFA>
.
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/yysijie/st-gcn/issues/81?email_source=notifications&email_token=AEOY3GFERBIP2HPQL6E25YTP2B3Y3A5CNFSM4FPOGCFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXPG24Y#issuecomment-501116275, or mute the thread https://github.com/notifications/unsubscribe-auth/AEOY3GCAO7F32NPNGE5LWP3P2B3Y3ANCNFSM4FPOGCFA .
-- ခင္စံပယ္ေထြး
— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.
Dataset used
Cooking Activity
Modified the data accordingly. The accuracies are not that great. I have tried changing the parameters multiple times, using pre trained models as well. Still no improvements. Also I get the same sequence of labels as output every time.
label_sequence = output.sum(dim=2).argmax(dim=0)Any idea where it might be going wrong?
Hi,@Zzoey ,there is a problem coming to me.Do you get the train_json file of your own dataset by the openpose directly?(e.g. you didn't add your custom post-processing function with the OpenPose C++/Python API ?) thx!
Hello, I have the skeleton points for each video but the input format is .npy. How could I generate .npy file from all my json files(openpose output)
Thanks in advance.
@yysijie how to get this file??thank you
Have you solved it，Can you help me
