Skip to content

Train tf lite object-detection model #68

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.

Sign up for GitHub

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

Jump to bottom
Open
kickermeister wants to merge 47 commits into
base: master
Choose a base branch
from
Open

Train tf lite object-detection model #68

kickermeister wants to merge 47 commits into from

Conversation

@kickermeister
Copy link
Collaborator

@kickermeister kickermeister commented Aug 31, 2021 •
edited
Loading

  • Create train, val, test database
  • Train model conveniently with command line interface and basic parameters
  • Convert fiftyoneImageDetection to expected format by tf make_model
  • upload / download files and dataset to cloud
  • create (balanced) random slices
  • and many more

To do:

  • import of trained model and class labels more conveniently
  • make the order of scores, bboxes and classes more reliable

@kickermeister
removed a few nonsense items from list, see also #54
6252a88
@kickermeister
added another list based on unique elements in food_testing_set.csv
d9fc2f8
@kickermeister
function added to convert fiftiyone metadata format to expected one b…
a339d0f 
…y tensorflow make_model
@kickermeister
fix: add additional empty columns at the end
6fcf313
@kickermeister
extended format conversion function and filter only real ingredients
a998c6f
@kickermeister
fix: tf make_model expects x_max / y_max and not width / heigth of bo…
efa72e2 
…unding box
@kickermeister
added a few functions to create dataset conveniently
2df347d
@kickermeister
some changes and imports for trainer.py
69cb492
@kickermeister
Merge branch 'tsflite_model' into get_training_data
2dae8d9
@kickermeister
added some initial basics to the trainer
29bf51c
@kickermeister
make tensorflow less verbose
b65f61d
@kickermeister
add missing dependency and make tf less verbose
a42f4af
@kickermeister
smaller classes list added
08d116f
@kickermeister
ingredient classes can be passed to functions
c2cea81
@kickermeister
fix: variable name
cc57c7c
@kickermeister
fix: use correct parameter name
3d59300
@kickermeister
fix parameter name
cf9b948
@kickermeister
fixed baseurl for images and model type
2aaf8f9
@kickermeister
fix path
ace1592
@kickermeister
add gcloud training command to Makefile
f4fdd39
@kickermeister
specify tensorflow version for gcloud AI platform training
f783000
@kickermeister
fix tf version
145dcd2
@kickermeister
function added to create randomly selected slice of datasets with bal…
55dde85 
…anced food classes
@kickermeister
fixed a few things and added __main__ part for gcloud training
451ea86
@kickermeister
changed a few names
13ead4a
@kickermeister
fix imports
14b753e
@kickermeister
Merge branch 'master' into get_training_data
9f3fd15
@kickermeister
use trained tflite model for metrics and API
99e9cb7
@kickermeister
enable a few more parameters for trainer
e1c4d00
@kickermeister
added the class labels map dictionary
562d2c7
@kickermeister
add function to load from datasets from cache
08bbe16
@kickermeister
fixed load from cache and enable automatic loading form local folder …
edfc6af 
…when folder exists
@kickermeister
fix condition
ecd6df0
@kickermeister
fix condition
091580d
@kickermeister
save class labels separately as pickle file
b30de1d
@kickermeister
enabled trainer parameters as cli arguments
dbbc9e3
@kickermeister
pass cli arguments to run method
b5cd3ef
@kickermeister
changed default model spec to lite0 again to avoid potential / likely…
1dd0a38 
… issues
@kickermeister
fix: add missing default value for spec
108f853
@kickermeister
fix parameter types
21e0476
@kickermeister
fix bool type action
a00954f
@kickermeister
revert bool action type
76a353a
@kickermeister
fix: filename of class labels
b4432cf
@kickermeister
fix export pickel name
12e7664
@kickermeister
Merge branch 'master' into get_training_data
def8cac
@kickermeister
use new model trained on ~40 classes with 16000 images. accuracy is n…
f656a27 
…ot very good. implemented import routine for class labels and export of those after training
@kickermeister
do not use predict_lite as default model
9031500
@kickermeister kickermeister changed the title WIP: Train tf lite object-detection model Train tf lite object-detection model Sep 2, 2021
@kickermeister
Copy link
Collaborator Author

kickermeister commented Sep 2, 2021

Would be ready for merging - even though we won't use the lite model in production

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@kickermeister