This time forum is not that compact 
updt1: Is there any chance somebody can move all Study group creation requests into one thread ?
updt2: And all ālooking forā into another single thread ?
updt3: Guys, lets create new topics wisely and use existing threads when its possible.
12 Likes
It doesnāt change anything for the rest (which was expected since itās a one pixel difference on the bboxes and just me being annoying), nothing bugs and the final loss is comparable. I was just wandering if there was some deeper thing there or if itās just a matter of convention.
Well thatās odd. I added it because without it, it was going outside the bounds of the image. I wonder what changed since then!
1 Like
When I saw it the first time, it seemed like that extra pixel was for the case where youād actually want to see the bbox clearly when the object coordinates are at the edges of the image.
Iām not sure that works, since each discussion needs to be separate. I think it would be better if someone created a google sheet or web app or something to create groups and have people join them. Then each group can create a group PM on the forum to arrange things directly, for instance. I wonāt have time to do so myself, but Iād be most grateful to anyone who creates useful tools to help folks in the course!
Following up on the how Part 2 is going focus a lot more on reading papers, is there a paper that anyone would recommend thatād be helpful and relevant for Lesson 8ās notebooks?
Thank you
1 Like
Guys, please check if we can move all requests for study groups into this Google sheet. All we need is to find if there is a group already in your Country/City and whom to contact to. If there is no group yet, just put yourself on the list.
Lets move all study groups discussions to private messages as there is no value in these threads for people not located in your lovely city.
8 Likes
Oh, I hadnāt noticed that those lower corners could reach the width/height of the image. It doesnāt cause any bugs in the plots though, and since we donāt try to access that pixel value (but maybe that happens later?) I didnāt get any error.
Will keep it in mind if a bug shows up as we continue to study that data set!
For those looking for bounding box generation tools for various platforms ā¦
See: https://en.wikipedia.org/wiki/List_of_manual_image_annotation_tools
6 Likes
At this point, we would say that if you are using a high end video card like the Titan X (Pascal) or possibly even a GTX 1080, it is probably a good idea to try to use a PCI-E 3.0 x16 slot - especially in multi GPU configurations.
This link reports the opposite of what Jeremy told us. But as far as I understand they only used gaming benchmarks ? Can this be applied to Deep Learning as well ?
There are links earlier in this thread to DL benchmarks showing x8 is fine.
Probably not the best way to deal with it, but Iām starting to āmuteā topics Iām sure that Iāll not be interested in. TMI otherwise.
2 Likes
Here is also good watch
https://www.google.com/search?q=Learning+greek+alphabet&oq=Learning+greek+alphabet&aqs=chrome..69i57j0l5.5786j0j7&sourceid=chrome&ie=UTF-8#kpvalbx=1
As @aza mentioned in class, the slightly prettier debugging interface include in IPython.core can be invoked via something like:
import IPython.core.debugger as pdb
Also, for using VSCode, PyCharm and other local editors remotely you may want to check out SSHFS. https://osxfuse.github.io/ if youāre using a mac.
11 Likes
Thereās also the IPython magic %pdb:
https://ipython.readthedocs.io/en/stable/interactive/magics.html#magic-pdb
Toggling it on will automatically activate pdb when a cell raises and exception.
2 Likes
Currently it is an unofficial support. Next release (0.4.0) of pytorch will be supporting Windows officially:
Knowing that the current version has been released on 12 Nov 2017, I hope the 0.4.0 release will be very soon.
Hi, I really like the part talking about software engineering skills and the pdb.
I have a question on pdb, if I accidentally run the cell again without quiting the debugger, it will hangs forever and interrupt/stopped does not help. Is there a way to solve this problem.
press n or s in debugger
there is no debugger, as I re-run the cell before I quit, the console disappear already so I cannot type anything.
When I use pdb.set_trace(), it will come up with a command line interface, but if I run the cell again before I quit the command line, notebook will freeze forever.
Congrats to the SF study group team for successfully completing their first project yesterday: fixing a tricky race condition bug in fastai.transforms: https://github.com/fastai/fastai/pull/234
7 Likes