sagieppel / Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation

Generator Evaluator selector net for image segmentation. Generator net generate Image segments. Evaluator net rank and select the best segments for the image segmentation map.
MIT License
10 stars 4 forks source link

Real Time #1

Open Daniel-1597 opened 5 years ago

Daniel-1597 commented 5 years ago

Hi,

Can it run in real-time?

sagieppel commented 5 years ago

No it take several second per image

On Fri, 8 Nov 2019, 13:55 Daniel-1597 notifications@github.com wrote:

Hi,

Can it run in real-time?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V55FVYVHE3SKOQ73BDDQSWYY7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HYBJ3UQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V57WOHAWM4HBA3LWJ5DQSWYY7ANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

Can we try to do it? Can you suggest me how to do it? Even bad fps would be fine.

On Sat, Nov 9, 2019, 1:55 AM sagieppel notifications@github.com wrote:

No it take several second per image

On Fri, 8 Nov 2019, 13:55 Daniel-1597 notifications@github.com wrote:

Hi,

Can it run in real-time?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub < https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V55FVYVHE3SKOQ73BDDQSWYY7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HYBJ3UQ , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AB37V57WOHAWM4HBA3LWJ5DQSWYY7ANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIJSZ2DXPSPP3D7C4LLQSXDKLA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDTIFGA#issuecomment-551977624, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMXRUILNIWYSNNENM2OBW3TQSXDKLANCNFSM4JK44RHQ .

sagieppel commented 5 years ago

There are many way the system can be accelerated perhaps even to real time. But I cant think of a way to do it without needing to rebuild and retrain the system.

On Fri, 8 Nov 2019, 23:01 Daniel-1597 notifications@github.com wrote:

Can we try to do it? Can you suggest me how to do it? Even bad fps would be fine.

On Sat, Nov 9, 2019, 1:55 AM sagieppel notifications@github.com wrote:

No it take several second per image

On Fri, 8 Nov 2019, 13:55 Daniel-1597 notifications@github.com wrote:

Hi,

Can it run in real-time?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V55FVYVHE3SKOQ73BDDQSWYY7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HYBJ3UQ

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AB37V57WOHAWM4HBA3LWJ5DQSWYY7ANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIJSZ2DXPSPP3D7C4LLQSXDKLA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDTIFGA#issuecomment-551977624 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AMXRUILNIWYSNNENM2OBW3TQSXDKLANCNFSM4JK44RHQ

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V575W7QARVQUM6GCEQTQSYY2DA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDT5FDY#issuecomment-552063631, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V57E4AI326LRXAAKNS3QSYY2DANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

I will also work on it if you guide me a bit is it possible for you?

On Sat, Nov 9, 2019, 7:55 PM sagieppel notifications@github.com wrote:

There are many way the system can be accelerated perhaps even to real time. But I cant think of a way to do it without needing to rebuild and retrain the system.

On Fri, 8 Nov 2019, 23:01 Daniel-1597 notifications@github.com wrote:

Can we try to do it? Can you suggest me how to do it? Even bad fps would be fine.

On Sat, Nov 9, 2019, 1:55 AM sagieppel notifications@github.com wrote:

No it take several second per image

On Fri, 8 Nov 2019, 13:55 Daniel-1597 notifications@github.com wrote:

Hi,

Can it run in real-time?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V55FVYVHE3SKOQ73BDDQSWYY7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HYBJ3UQ

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AB37V57WOHAWM4HBA3LWJ5DQSWYY7ANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIJSZ2DXPSPP3D7C4LLQSXDKLA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDTIFGA#issuecomment-551977624

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AMXRUILNIWYSNNENM2OBW3TQSXDKLANCNFSM4JK44RHQ

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub < https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V575W7QARVQUM6GCEQTQSYY2DA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDT5FDY#issuecomment-552063631 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AB37V57E4AI326LRXAAKNS3QSYY2DANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIK6UYMZ2GNQ3FMMGSLQS3B6ZA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUHKCA#issuecomment-552105224, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMXRUIKZIPRAT2MJKSCBAOLQS3B6ZANCNFSM4JK44RHQ .

sagieppel commented 5 years ago

Sure will be happy to help. If you do it for the research or have specific property that you need like class agnostic segmentation then that good.

But if you just looking for panoptic segmentation with real time performance. Then there are other methods that already do this that can just download.

On Sat, 9 Nov 2019, 10:25 Daniel-1597 notifications@github.com wrote:

I will also work on it if you guide me a bit is it possible for you?

On Sat, Nov 9, 2019, 7:55 PM sagieppel notifications@github.com wrote:

There are many way the system can be accelerated perhaps even to real time. But I cant think of a way to do it without needing to rebuild and retrain the system.

On Fri, 8 Nov 2019, 23:01 Daniel-1597 notifications@github.com wrote:

Can we try to do it? Can you suggest me how to do it? Even bad fps would be fine.

On Sat, Nov 9, 2019, 1:55 AM sagieppel notifications@github.com wrote:

No it take several second per image

On Fri, 8 Nov 2019, 13:55 Daniel-1597 notifications@github.com wrote:

Hi,

Can it run in real-time?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V55FVYVHE3SKOQ73BDDQSWYY7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HYBJ3UQ

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AB37V57WOHAWM4HBA3LWJ5DQSWYY7ANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIJSZ2DXPSPP3D7C4LLQSXDKLA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDTIFGA#issuecomment-551977624

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AMXRUILNIWYSNNENM2OBW3TQSXDKLANCNFSM4JK44RHQ

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub <

https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V575W7QARVQUM6GCEQTQSYY2DA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDT5FDY#issuecomment-552063631

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AB37V57E4AI326LRXAAKNS3QSYY2DANCNFSM4JK44RHQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AMXRUIK6UYMZ2GNQ3FMMGSLQS3B6ZA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUHKCA#issuecomment-552105224 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AMXRUIKZIPRAT2MJKSCBAOLQS3B6ZANCNFSM4JK44RHQ

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V56KAW4NXVWVLHP6JOLQS3I7ZA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUIPSQ#issuecomment-552110026, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V54BSGETY2E6MLGLXBTQS3I7ZANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

I am looking for panoptic segmentation can you tell me what are the other repos

sagieppel commented 5 years ago

Ups net for exanple

On Sat, 9 Nov 2019, 11:06 Daniel-1597 notifications@github.com wrote:

I am looking for panoptic segmentation can you tell me what are the other repos

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V5Z4GLZXODBD7BZM5S3QS3NZTA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUJJTQ#issuecomment-552113358, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V56OQGGJL5UQQQ2C2PDQS3NZTANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

That is not working there are some issues with the repo

sagieppel commented 5 years ago

What are you trying to do and what is the error message?

On Sat, 9 Nov 2019, 21:59 Daniel-1597 notifications@github.com wrote:

That is not working there are some issues with the repo

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V52T4BF5UEWNR3OKNQTQS52IVA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDUUC4Y#issuecomment-552157555, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V533H2XDRXYDYUQRZ73QS52IVANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

Sir, can you write code for a video?

sagieppel commented 5 years ago

If you have a problem with the ups net repo post them in the ups net repo.

On Sun, Nov 10, 2019 at 11:05 PM Daniel-1597 notifications@github.com wrote:

[image: Screenshot from 2019-11-11 09-34-19] https://user-images.githubusercontent.com/53418529/68560287-75568080-0466-11ea-84ed-517b17c5ef15.png This is the error I am facing and I couldn't move further. and files are present in the repo

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V52NQBYLCTU53CMI6YTQTDKX3A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDVTHCY#issuecomment-552285067, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V574JNYNTG4IP6FVJITQTDKX3ANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

sure sir Can you write code for a video also?

Daniel-1597 commented 5 years ago

Sir when I am trying to run the classification test.py file what is the location of the mask directory?

sagieppel commented 5 years ago

what file do you refer to exactly?

On Mon, Nov 11, 2019 at 1:45 AM Daniel-1597 notifications@github.com wrote:

Sir when I am trying to run the classification test.py file what is the location of the mask directory?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V526DRYGY46MUJOW3JLQTD5P7A5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDV2R5Q#issuecomment-552315126, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V53I6C3N455V3JN25YDQTD5P7ANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

Screenshot from 2019-11-11 13-28-59

Daniel-1597 commented 5 years ago

in this file please change Augment to AugmentImage= False

sagieppel commented 5 years ago

Ok your mask image dir paths point to folder in my computer so their clearly different for you. Why do even use need this script?

On Mon, 11 Nov 2019, 3:00 Daniel-1597 notifications@github.com wrote:

in this file please change Augment to AugmentImage= False

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V54BOP7C7MD5FHZECBDQTEGIDA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDV7EFA#issuecomment-552333844, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V55CE4F5FPMNTCXGWALQTEGIDANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

I just want to know the classes in the image that's why

sagieppel commented 5 years ago

Just running the run.py in the main folder will give you annotation maps with classes according to the coco panoptic format. See readme for details.

On Mon, 11 Nov 2019, 4:43 Daniel-1597 notifications@github.com wrote:

I just want to know the classes in the image that's why

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V5ZWMW53C3HQO6K4KRTQTESLHA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDWHP7Q#issuecomment-552368126, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V52D2WUOVYL6AKZDOY3QTESLHANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

Sir, How can we optimize the GESModularClass.py. Any suggestions for running it in real-time? I have asked the same question so many times can please answer clearly so that I can modify or write a new code for real-time. Thank You, Daniel.

sagieppel commented 5 years ago

You can accelerate the net by unifying the four modular nets into one net and by merging the roi and pointer point mask in higher layers. That will lead to signifcant acceleration. But its a lot of work and demand considerable understanding of nets. If you just need fast panoptic segmentation just download another model like the ups net

On Tue, 12 Nov 2019, 23:13 Daniel-1597 notifications@github.com wrote:

Sir, How can we optimize the GESModularClass.py. Any suggestions for running it in real-time? I have asked the same question so many times can please answer clearly so that I can modify or write a new code for real-time. Thank You, Daniel.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V53GNCO6XAO4WG6WLWTQTN5GJA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOED42ECQ#issuecomment-553230858, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V52YKWCMBXUTV2LPBL3QTN5GJANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

The results are very good and the documentation of your repo is very good. If I consider UPS net the documentation is very bad and I can't even run it. I want to work more on your repo and make real-time (even if it is giving bad fps). The work you have done and documentation is Awesome but yes it's too much computation and for training also it will take a good amount of time.

You are using four networks to achieve panoptic segmentation which is consuming a lot of time.

I have an Idea is it possible to make it into a single model? If it can happen in one network I think this may be faster

sagieppel commented 5 years ago

Well running working existing net even with bad dicumentation is much much easier then rebuilding and retraining the nets from scratch. For your time I still advice you against it. But if you want to give it a try anyway go for it.

On Wed, 13 Nov 2019, 3:34 Daniel-1597 notifications@github.com wrote:

The results are very good and the documentation of your repo is very good. If I consider UPS net the documentation is very bad and I can't even run it. I want to work more on your repo and make real-time (even if it is giving bad fps). The work you have done and documentation is Awesome but yes it's too much computation and for training also it will take a good amount of time.

You are using four networks to achieve panoptic segmentation which is consuming a lot of time.

I have an Idea is it possible to make it into a single model? If it can happen in one network I think this may be faster

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/sagieppel/Generator-evaluator-selector-net-a-modular-approach-for-panoptic-segmentation/issues/1?email_source=notifications&email_token=AB37V52BR6IIFX3QV47AX5TQTO33BA5CNFSM4JK44RH2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOED5KAFI#issuecomment-553295893, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37V54BNF3MABTE4VTUA4TQTO33BANCNFSM4JK44RHQ .

Daniel-1597 commented 5 years ago

I accept with you building a model is very difficult from scratch the issue I am facing Import errors with upsnet