SSD EfficientNet in tensorflow keras version

Unofficial implementation of SSD with EfficientNet backbone using tf keras.

UPDATE

Here is the link for official EfficientNet implementation for TPU training.

Install Requirement

• Python 3.X
• TensorFlow 1.13.1 (Use current version 1.14 would cause a wired dimension mismatch caused by se-expand operation in se block.)
• TensorFlow Models master branch
• Protoc 3.5.7

Usage Steps

1. Put efficientnet.py and efficient_feature_extractor.py under object_detection/models directory
2. Modify model_builder.py and add SSDEfficientNetFeatureExtractor and SSDEfficientNetFPNFeatureExtractor

   
   from object_detection.models.efficientnet_feature_extractor import SSDEfficientNetFeatureExtractor, SSDEfficientNetFPNFeatureExtractor

SSD_KERAS_FEATURE_EXTRACTOR_CLASS_MAP = { ... 'ssd_efficientnet': SSDEfficientNetFeatureExtractor, 'ssd_efficientnet_fpn': SSDEfficientNetFPNFeatureExtractor, }

3. Replace `ssd.proto` file under `protos` with this one. **Then make sure to rerun `protoc object_detection/protos/ssd.proto --python_out=.`**
4. Install TensorFlow object detection api: see [here](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md)
5. Train model following [official steps](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/running_locally.md)
6. Refer to `ssd_efficientnet.config` and `ssd_efficientnet_fpn.config` as example config (If use `ssd_efficientnet_fpn` backbone, make sure `use_explicit_padding` is `true` to prevent potential mismatch error during top down addition operation).