RAM leak problem when training in command batch_queue.get(). Where do you release resources once a batch is trained. about maskflownet

[vivasvan1]

Can you check if it is only on my pc or is this happening with your code too?

also, is there any way in which I don't have to load the full dataset on the memory for training in mxnet?

[vivasvan1]

I have found using pdb that after every run of

batch = batch_queue.get()

an extra 0.10-0.15% ram is consumed which seems to never get released.

(Pdb) print("| ram=",psutil.virtual_memory().available * 100 / psutil.virtual_memory().total)
 **ram= 28.66921519345203** 
(Pdb) n
> /home/mask/maskflownet/MaskFlownet/main.py(572)<module>()
-> loading_time.update(default_timer() - t0)
(Pdb) print("| ram=",psutil.virtual_memory().available * 100 / psutil.virtual_memory().total)
**ram= 28.542640291935687**

I cannot find why this is happening but i am sure of it. Can you help me fix this please?

[simon1727]

Hi vivasvan1, thanks for pointing out this problem.

We import this Queue method from the python queue package directly without any modification. I search on Google and find that other people encounter the same problem. So maybe this is not a problem with our code but the python queue package.