thatwist/p24-lions

p24-lions

Link to Trello board

Software / Stack

• Python3
• OpenCV 3.4
• GCP with GPUs

Dataset

1. https://www.albert.cm/projects/viewpoint_3d_pose/
2. http://vrai.dii.univpm.it/re-id-dataset
3. https://www.albert.cm/projects/ram_person_id/

dataset/ - cut heads dataset

Final pipeline

final-take2/

Presentation

https://docs.google.com/presentation/d/14u4VbT3fY137V4mZnPOO8NkiWFpT-450tGD39dheFoE/edit?usp=sharing

Tensorflow Object Detection API

Tutorial

Installation

• Dataset - https://www.albert.cm/projects/viewpoint_3d_pose/

Generating TFRecords

python generate_tfrecord.py --csv_input=data/train.csv --output_path=data/train.record

Preparing data

1. download and put training / testing data into depth-train/data (h5 files for depth-map, labels for train/test - 4 files)
2. generate csv files using generate_csv_labels function from prepare-dataset.ipynb
3. generate tfrecord files from csvs using generate_tfrecord.py (see above)
4. follow tutorial to prepare config files

Run training

python legacy/train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_coco.config

Run eval

python legacy/eval.py --logtostderr --pipeline_config_path=training/ssd_mobilenet_v1_coco.config --checkpoint_dir=training/ --eval_dir=eval/

eval parameters: https://github.com/tensorflow/models/blob/abd504235f3c2eed891571d62f0a424e54a2dabc/research/object_detection/protos/eval.proto#L8

Run tensorboard monitoring (for evaluation)

tensorboard --logdir=eval

Export model

python export_inference_graph.py --input_type image_tensor --pipeline_config_path=training/ssd_mobilenet_v1_coco.config --trained_checkpoint_prefix=training/model.ckpt-400 --output_directory output

Python

• locale problem

export LC_ALL="en_US.UTF-8"
export LC_CTYPE="en_US.UTF-8"
sudo dpkg-reconfigure locales