How to view progress on Tensorboard

[SamihaSara]

I tried to see progress on Tensorboard ( logs at SiamMask/experiments/siammasm_base/logs/) with the following command by port forwarding from server to local pc.

ssh -L 16006:127.0.0.1:6006 user@host Then you run the tensorboard command: tensorboard --logdir=/path/to/logs then tried to access the tensorboard in browser under: localhost:16006/

But it did not output any graph or update. If someone has completed this step, please help.

[cnctdadots]

%tensorboard --logdir [directory name of the log files]

[SamihaSara]

@tedrostime Hi, thanks for helping. I have tried this also but got this error-- -bash: fg: %tensorboard: no such job. Also, I am using 2 GPUs where the paper uses 4. Can you suggest which parameters should be modified for this?

[cnctdadots]

I think you included the '%' symbol. you shouldn't include the % symbol. Just 'tensorflow --logdir [your directory name]' works

[SamihaSara]

I tried 'tensorflow --logdir [your directory name]', got this output: TensorBoard 1.13.1 at http://devcube4:6007 (Press CTRL+C to quit) But how to visualize this link? I have tried it on chrome in pc, it shows nothing.

[cnctdadots]

the directory name is saved in 'board'. you can check that in tools/train_siammask.py (line # 63) so the tensorboard will be as: tensorboard --logdir board/ (something like that)

[SamihaSara]

Just tried this. Again nothing in the browser. Can you also tell, how should I open the link -- http://devcube4:6007?

[cnctdadots]

Just tried this. Again nothing in the browser. Can you also tell, how should I open the link -- http://devcube4:6007?

I think copy and paste the link to your browser.

as a last option, you can try this: tensorboard --logdir=board/ --host localhost --port 8088 (make sure your path to the logdir is the correct)

[cnctdadots]

I think this will work now. copy & paste the link given (http://localhost:8089) to your browser. hope that works

[SamihaSara]

Sorry, It did not work again.

[cnctdadots]

Sorry, I did not work again.

in the browser, you are writing 8088, but from the terminal it is in port 8089. can you check that part?

[SamihaSara]

Sorry I provided the wrong picture.

[cnctdadots]

I don't know why but can you try this again: tensorboard --logdir board/

[SamihaSara]

@mesmertrack tried, does not work. Just to confirm, Am I supposed to open the link in the local machine's browser? Is there any way to open this link on the server?

[cnctdadots]

oh wow, I have no idea why it is not working. hope someone else helps in this. As your question, I think since you are training your dataset in your local machine, opening the link in local browser makes sense. Good luck

[SamihaSara]

@mesmertrack no I am training the model on a server. Can you tell me if I can open the link there?

[cnctdadots]

I am not quite sure but this link may help. https://stackoverflow.com/questions/38464559/how-to-locally-view-tensorboard-of-remote-server/49101487

[SamihaSara]

@mesmertrack Thank you so much. One of the solutions from this link worked. Here it is

ssh to your server and find its IP via terminal by running: IP=hostname -I Open the tensorboard server on the host server: tensorboard --logdir=board --host $IP Use your browser to and surf to http://$IP:6006

[cnctdadots]

great. glad it worked. good luck on your works!

[SamihaSara]

@mesmertrack can you please provide some info to debug this by mentioning which parts of the project contain codes for these tasks: Mask representation, ROWs refinement, and Box generation?

[cnctdadots]

@SamihaSara : if you check the tools/train_siammask.py, it has train(...) function. you can check the working flow from there. or to make it simple, start from the main() function.

[SamihaSara]

@cnctdadots Can you please tell where and why in the SiamMask architecture, SiamRPN is used? I have read the paper but could not figure out.