RetinaNet tensorflow version

Unofficial realization of retinanet using tf. NOTE this project is written for practice, so please don't hesitate to report an issue if you find something run.

TF models object detection api have integrated FPN in this framework, and ssd_resnet50_v1_fpn is the synonym of RetinaNet. You could dig into ssd_resnet50_v1_feature_extractor in models for coding details.

Since this work depends on tf in the beginning, I keep only retinanet backbone, loss and customed retinanet_feature_extractor in standard format. To make it work, here are the steps:

• Download tensorflow models and install object detection api following this way.

• Add retinanet feature extractor to model_builder.py:

from object_detection.models.retinanet_feature_extractor import RetinaNet50FeatureExtractor, RetinaNet101FeatureExtractorSSD_FEATURE_EXTRACTOR_CLASS_MAP = {    ...
    'retinanet_50': RetinaNet50FeatureExtractor,    'retinanet_101': RetinaNet101FeatureExtractor,
}

• Put retinanet_feature_extractor.py and retinanet.py under models

• Modify retinanet_50_train.config and train.sh with your customed settings and data inputs. Then run train.sh to start training.