-
Notifications
You must be signed in to change notification settings - Fork 154
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
TFLite SSD model with post processing. #247
Conversation
Archive is 26 MB. Do you need all files from it? Can you shrink its size by removing unnecessary files? |
There is only file inside - Just TFlite model |
is it possible to manually construct a tflite model that is smaller? e.g. do a graph cut to cut off the previous layers and only include the NMS part |
I don't have experience with TF graph transforms, but it might be possible. Currently, there is accuracy mismatch so we cant replace the older model. I was hoping to add this model to open an Issue so that the original authors can take a look. |
yes, old file |
@tqchen apache/tvm#5439 is now passing, so I have removed the older TFLite file. Let me know if this is ok. |
Just realized that merging this will cause failure in TVM master as the current tests are dependent on the model (*_nopp.tgz) that I am removing in this PR. I can send a separate TVM PR that just deletes the test using *_nopp.tgz @tqchen Wanted to check if this makes sense? |
@tqchen Let me know if you need any more changes here. The TVM PR to remove old test is merged. |
Ping @tqchen :) |
@apivovarov @tqchen
I will add a test case in TVM for TFLite frontend. There is NMS support but not tested through TFLite frontend.
The instructions are exactly same as that of
#186
But with post processing enabled. Thanks @apivovarov for providing the instructions.