Skip to content

Latest commit

 

History

History
32 lines (23 loc) · 2.4 KB

README.md

File metadata and controls

32 lines (23 loc) · 2.4 KB
Raw

All credit for this AI model goes to Anusua Trivedi, Senior Data Scientist for Microsoft AI for Good Team. The AI model is Copyright (c) Microsoft Corporation. All rights reserved.

TO GET Tensorflow API:

  1. Clone github: https://github.com/tensorflow/models
  2. Use code here: https://github.com/tensorflow/models/tree/master/research/object_detection

TO ADD TF API TO PATH:

export PYTHONPATH=$PYTHONPATH:/home//<base_dir>/models/research:/home/antriv/notebooks/models/research/slim

TO TEST TF API: (run from the reseach folder)

python home//<base_dir>/models/research/object_detection/builders/model_builder_test.py

TO PREPROCESS images/annotations:

  1. python /home//<base_dir>/nvidia_smoke/data/smoke_annotations/train_val_test_split.py
  2. python /home//<base_dir>/nvidia_smoke/data/smoke_annotations/generate_tfrecord.py --csv_input=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/train.csv --output_path=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/train.record --image_dir=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/PNGImages
  3. python /home//<base_dir>/nvidia_smoke/data/smoke_annotations/generate_tfrecord.py --csv_input=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/test.csv --output_path=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/test.record --image_dir=/home//<base_dir>/nvidia_smoke/data/smoke_annotations/PNGImages

TO TRAIN USING TF API:

python /home//<base_dir>/models/research/object_detection/model_main.py
--pipeline_config_path=/home//<base_dir>/nvidia_smoke/ssd_mobilenet_v2_smoke/pipeline.config
--model_dir=/home//<base_dir>/nvidia_smoke/ssd_mobilenet_v2_smoke/smoke_model
--num_train_steps=10000
--alsologtostderr

TO FREEZE THE TRAINED TF MODEL:

python /home//<base_dir>/models/research/object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path /home//<base_dir>/nvidia_smoke/ssd_mobilenet_v2_smoke/pipeline.config --trained_checkpoint_prefix /home//<base_dir>/nvidia_smoke/ssd_mobilenet_v2_smoke/smoke_model/model.ckpt-10000 --output_directory /home//<base_dir>/nvidia_smoke/ssd_mobilenet_v2_smoke/smoke_model/frozen_graph

TO TEST USING TF API:

python /home//<base_dir>/models/research/object_detection/nvidia_smoke_OD_inference.py NOTE: You will need Xming server client to test locally