Skip to content

l3p-cv/lost_cls_keras

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LOST cls keras

Description

This Repo is used to train a Tensorflow Keras model to use it with lost_semi_auto_pipes Repo. The description and the default options are only to train a Tensorflow Keras "MobilNet" model. To train an other model look at Keras.

Installation

First, clone or download this GitHub repository and install requirements.

pip install -r ./requirements.txt

Data

To train the model you need:

  • annotation data as parquet file

You get annotation data after the first iteration of "request_triton_mia_loop" pipeline.

directory structure:

.
├── Images
├── anno_data
│   ├── anno_data_0.parquet
│   ├── anno_data_1.parquet
│   └── lbl_map.json
├── checkpoints
│   ├── checkpoint
│   ├── model_name.data-00000-of-00001
│   └── model_name.index
├── LICENSE
├── CHANGELOG.md
├── config.py
├── README.md
├── requirements.txt
└── train.py

Edit the configs

To train your model, edit the config.py

configs.py

  • line 2: If it is the first training of the model set it false else true
  • line 3: Name of the model.
  • line 4: Path of the image dataset
  • line 5: Path of annotation data parquet file. Create this folder for all your annotation parquet files.
  • line 7: Image size to scale for training. MobilNet model supports 128, 160, 192 and 224
  • line 8: Choose image batch size per training epoch.
  • line 10: Choose the number of training epochs.
  • line 13: Path of your model repo.

Training process

  1. download anno_data.parquet from data export of "request_triton_mia_loop" pipeline
  2. load or add your anno_data.parquet into your created directory
  3. run train.py
    • train the model and save the weights for best model at the checkpoints directory
    • save the model as a TensorFlow model
    • check the model directory for older versions
    • save the new model version with one counter higher as the version befor

About

Train classification models for lost

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages