LogoDetection

Logo Detection using Tensorflow (2) Object Detection API as part of the Final Project for Bocconi's Deep Learning Course.

All the steps presented here can be found in the collab_tutorial section. The pipelines for training were also upload.

Intro

In social media analysis, sometimes images/videos can inform us more than the pure text. For example, if you want to do a search on users of a certain brand, it is perhaps more effective to look for images/videos, as it is unlikely that anyone will explicitly write the company in their post. Therefore, combined with textual analyses, this methodology can improve the result of such analyses.

Results

Coco metrics, mAP 0.5 = 0.963 and mAP 0.5:0.95 = 0.761.

Workflow

Object Detection API

Instaling Objet Detection API using Phyton in a Linux Machine. More information can be found here.

# Clone git repository
git clone https://github.com/tensorflow/models.git
cd models/research
# Compile protos.
protoc object_detection/protos/*.proto --python_out=.
# Install TensorFlow Object Detection API.
cp object_detection/packages/tf2/setup.py .
python -m pip install --use-feature=2020-resolver .

Dataset

The dataset used for training was scrapped from public publications on instagran between 2015 and 2017 in some US cities. The logos used for training are: Nike, Adidas, Under Armour, Puma, The North Face, Starbucks, Apple Inc., Mercedes-Benz, NFL, Coca-Cola, Chanel, Toyota, Pepsi, Hard Rock Cafè.

Generate TFRecords

TFRecords were created using the official documentation.Some parts of the code in colab was inspired in the racoon dataset.

Pre-trained COCO models

Pre-trained coco models can be found here. For this project, it was used the CenterNet HourGlass104 512x512.

$ wget http://download.tensorflow.org/models/object_detection/tf2/20200713/centernet_hg104_512x512_coco17_tpu-8.tar.gz

Recoment Data Structure

.
- data/
│   ├── eval-00000-of-00001.tfrecord
│   ├── label_map.txt
│   ├── train-00000-of-00002.tfrecord
│   └── train-00001-of-00002.tfrecord
└── models/
    └── my_model_dir/
        ├── eval/                 # Created by evaluation job.
        ├── my_model.config
        └── model_ckpt-100-data@1 #
        └── model_ckpt-100-index  # Created by training job.
        └── checkpoint            #

Training

The official documentation for training and evaluation with Object Detection API, can be found here

$ pipeline_file = PATH/TO/PIPELINE_FILE
$ model_dir = PATH/TO/MODEL_DIR
$ steps = 50000
$ train_file = PATH/TO/model_main_tf2.py
$ python ${train_file} \
    --pipeline_config_path={pipeline_file} \
    --model_dir={model_dir} \
    --alsologtostderr \
    --num_train_steps={steps} \
    --sample_1_of_n_eval_examples=1 

Export Graph

$ output_directory = PATH/TO/EXPORT_GRAPH
$ python ${train_file} \
    --trained_checkpoint_dir {model_dir} \
    --output_directory {output_directory} \
    --pipeline_config_path {pipeline_file}

Evaluation

$ python ${train_file} \
    --pipeline_config_path=${pipeline_file} \
    --model_dir=${model_dir} \
    --checkpoint_dir=${last_model_path} \
    --alsologtostderr

IOU Results

LOGO	IOU Header
THE NORTH FACE	0.920
STARBUCKS	0.919
TOYOTA	0.908
MERCEDES-BENZ	0.905
UNDER ARMOUR	0.899
ADIDAS	0.898
APPLE INC.	0.894
HARD ROCK CAFE	0.894
PUMA	0.873
NFL	0.868
NIKE	0.868
CHANEL	0.861
PEPSI	0.838
COCA-COLA	0.838

License

MIT