aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-461618966
I put the tar files in a separate bucket so you can pick how you want to
download.
To get the tar files:
```
aws s3 sync
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-461254738
@thomelane Ok, I'm creating the tar files now. The reason I didn't do that
is that I kept getting disconnected and wanted to be able to resume
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-457411369
@seujung @soeque1 Let me know when you're ready for another review.
This is
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-456561421
I see one [outstanding suggestion to put in the Python3
flag](https://github.com/apache/incubator-mxnet/pull/13647#discussion_r248661689)
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-453380303
I made it public read. You need AWS cli installed. 'sudo apt install awscli'
You might need to still create an account with IAM and run aws
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-453211484
Putting this here, so when my comment in the review gets outdated, you still
can see how to get the preprocessed data:
The readme could
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-452475997
Multi-gpu training is working! Is there any advice you want to provide in
the Readme about recommended maximum batch size or how to scale up
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-452405729
Thanks for adding multi-gpu - that's great! I'll test it out.
Restarting CI due to a "OSError: [Errno 14] Bad address" error. Probably
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-452027678
The command line example seems to be working. It would be a nice bonus if
this also used multi-gpu. Or is that an option that I missed?
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-451636856
I ran into another bug that is easily resolved.
```
mxnet.base.MXNetError: [17:39:13] src/io/local_filesys.cc:199: Check failed:
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-451452405
With regard to the data that is preprocessed - is this just the /data/align
folder that is needed? So if I upload that to s3 that is all that
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-451425540
> * (3-2) Stop the preprocess. Then resume the process. If there is the
processed image, it skips.
This worked out! I finished
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-451068779
> When you did run 'preprocess_data.py', it actually did nothing because of
bug.
> I changed the paths with bug. it takes about 6 hours
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-450309197
I built the project on a GPU instance this time and was able to run main.py.
However, I immediately get a dump of a lot of these errors:
aaronmarkham commented on issue #13647: Update lip reading example
URL: https://github.com/apache/incubator-mxnet/pull/13647#issuecomment-450302412
Please add these to your prerequisites list:
* scikit-image
* scikit-video
* dlib
Also, I tried to run it without a GPU and
15 matches
Mail list logo