corn

Predict corn.

Members

Zachary Balda
Brandon Boynton
Laurent Valle

Data

.
+-- data
|   +-- GOPRO165_01.jpg
|   +-- GOPRO165_01.json
|   +-- GOPRO165_02.jpg
|   +-- GOPRO165_02.json
|       ...
|   +-- GOPRO165.MP4
|   +-- GOPRO166_01.jpg
|   +-- GOPRO166_01.json
|       ...
|   +-- GOPRO166_18.jpg
|   +-- GOPRO166_18.json
|   +-- GOPRO166.MP4
|       ...
|       ...

How to Train

At path /keras-frcnn/ run:

python train_frcnn.py -o simple -p annotations.txt

Note: replace annotations.txt with actual annotations file

How to Inference

Place images in /keras-frcnn/test_images/

At path /keras-frcnn/ run:

python test_frcnn.py -p test_images

Outputs are saved to /keras-frcnn/results_images/

Note: Images must be .jpg format

Converting MP4 Videos to .jpg Image Sequences

The keras-frcnn model only accepts .jpg images. To convert MP4 videos to .jpg image sequences put all MP4 videos in /data/ and run the script we created:

python mp4_to_jpg.py

Set fps in mp4_to_jpg.py to choose number of frames per second (default 1)

Converting JSON Annotations to .txt File

Annotations were done with labelme which saves a json file for each annotated image (at /data/). The keras-frcnn model accepts a single .txt file which has a specific format. To generate a single .txt file from the json files, navigate to /data/ and run the script we created:

python json_to_txt.py

Dependencies

pandas
matplotlib
tensorflow
keras – 2.0.3
numpy
opencv-python
sklearn
h5py

zbalda/corn