mobilenetv2-yolov3

Tensorflow implementation mobilenetv2-yolov3 inspired by keras-yolo3

Update

Backend:

• MobilenetV2
• Efficientnet
• Darknet53

Callback:

• mAP
• Tensorboard extern callback

Loss:

• MSE
• GIOU

Train:

• Multi scale image size
• Cosine learning rate

Tensorflow:

• Tensorflow2 Ready
• Use tf.data to load dataset
• Use tfds to load dataset
• Remove image shape input when use session
• Convert model to tensorflow lite model
• Multi GPU training support

Serving:

• Tensorflow Serving warm up request
• Tensorflow Serving JAVA Client
• Tensorflow Serving Python Client
• Tensorflow Serving Service Control Client
• Tensorflow Serving Server Build and Plugins develop

Requirement

Tensorflow-1.14+

Numpy-1.16.2+

Python-3.6.7+

Usage

Change arguments in main.py,then exec

python main.py

---

Train

• Download pascal tfrecords from here.
• Change train.py

opt = <your session config>
backbone = <your yolov3 backbone>
log_dir = <path/to/your/tensorboard/log>
batch_size = <you batch size>
train_dataset_path = <path/to/your/train/folder>
val_dataset_path = <path/to/your/val/folder>
train_dataset_glob = <train glob>
val_dataset_glob = <val glob>

---

Performance

3 times faster than darknet53-yolov3 with alpha=1.4 and higher accuracy

Pascal Dataset

I have packaged a pascal tfrecords for you.See here

Reference:

paper:

• YOLOv3: An Incremental Improvement
  
• Foca Loss for Dense Object Detection
  
• Group Normalization
  
• An Analysis of Scale Invariance in Object Detection - SNIP
  
• Scale-Aware Trident Networks for Object Detection
  
• Understanding the Effective Receptive Field in Deep Convolutional Neural Networks
  
• Bag of Freebies for Training Object Detection Neural Networks
  
• Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regression
  
• MobileNetV2: Inverted Residuals and Linear Bottlenecks