Project Description

Inference results for USA downstreet video streaming

Feature extraction and measurement of butterflies and orchids

• Extraction for average blackness of butterfly eyespot through Mask RCNN.

• Extraction of orchid featuremap from different layers of Residual Network (ResNet).

• Extraction of orchid featuremap from different layers of Feature Pyramid Network (FPN).

• The Average-Precision (AP) and training time for different Mask RCNN models (orchids).

Extraction for orchid root length and leaf width using skeleton extraction algorithm and min-rectangle area method via output masks

• Skeleton extraction algorithm (image morphology).

• Min-rectangle area method.

Inference results

Different values for the features of orchids measured via output boundingboxes, masks, and classes

Setup & run code

Getting started

• Clone this repo to your local

git clone https://github.com/simonyang0608/Mask_RCNN_Project
cd Mask_RCNN_Project

Computer equipments

• System: Ubuntu16.04
• Python version: Python 3.5 or higher
• Keras version: Keras 1.8.4
• Tensorflow version: Tensorflow 1.8.0
• Training:
  CPU: Intel Xeon E5-2698 Dual 20 cores @ 42.2 GHz
  GPU: NVIDIA Tesla V100 16GB*8
• Testing & inference:
  CPU: Intel(R) Core(TM)i5-7300HQ CPU @ 2.5.0 GHz
  GPU: NVIDIA GeForce GTX 1050 2GB

Training

• You should download the released pretrained models. And put the model on the folder ./mrcnn_model to train related models continously.

python3 mrcnn_model/parallel_model.py

Testing & inference

• You should download the released pretrained models. And put the model on the folder ./mrcnn_model to inference the test set on the folder ./pic.

• Orchids dataset or video streaming

python3 mask_rcnn_create_live_stream.py

• Butterfly dataset

python3 mask_rcnn_detect_butterfly.py

To get the released pretrained model

https://github.com/simonyang0608/Mask_RCNN_Project/releases/tag/1