object_detection

This repository is forked and modified from tensorflow object detection model.

To run baseline code for the hackathon, you may follow the instruction below:

1. fork / mirror this repository

   !!!important: make it as your own private repository and assign right to YITU admin （refer to section 5 in "Hackathon_Baseline_User_Guider"!!!

2. environment setttings

   # activate AWS virtual environment
   cd ~
   source activate tensorflow_p36

   # Setup PYTHONPATH
   cd hackathon-sg/ 
   export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim    

3. to generate tfrecord data

   cd object_detection
   python script/generate_tfrecord_from_csv.py --image_dir ../input/training/images/ --output_path ../input/hackathon --csv_file ../input/training/train_label.csv --validation_set_size 500

4. to train the model

   python script/train.py --logtostderr --train_dir=training/baseline/ --pipeline_config_path=training/hackathon_baseline.config

   #to visualize the training results, can only use 8000 for port no. tensorboard --logdir=training/baseline --port 8000

5. to eval the trained model

   python script/eval.py --logtostderr --pipeline_config_path=training/hackathon_baseline.config --checkpoint_dir=training/baseline --eval_dir=eval/baseline

   #To visualize the eval results, can only use 8000 for port no. tensorboard --logdir=eval/baseline --port 8000

6. to export the trained model

   python script/export_inference_graph.py --input_type image_tensor --pipeline_config_path training/hackathon_baseline.config   --trained_checkpoint_prefix training/baseline/model.ckpt-20000 --output_directory output/

7. to output the results to .csv

   python script/output_csv_results.py threshold=0.5 data_dir=../input/testing/images/ model_path=output/frozen_inference_graph.pb output_path=output/submission.csv label_map=../input/label_map.pbtxt

8. to submit the results

   please follow the instruction from ”Hackathon Infra User Manual“ to submit your detection results.

   git push your modified codes to your private repository whenever you make a submission.

   !!!DO PUSH YOUR CODES FOR EVERY SUBMISSION YOU MADE FOR CODE VERIFICATION PURPOSE!!!