leonkuhne / ladys-slipper-counter-ai

Using YOLO and RE3 libraries to count lady's slippers orchids in drone video data.

Check out https://github.com/thtrieu/darkflow

This repo is an in-development research effort to count ladies slipper orchids from drone footage for the purpose of monitoring populations.

---

First Time Setup

Open up a terminal, cd to this directory, run:

make

Create Directories

sample_img/

contains validation images for run.py

bin/

yolo weights

annotations/

images/

Download YOLO weights

These can be found (link somewhere) on the Darkflow GitHub Add the downloaded weights to 'bin' and rename to 'yolo.weights'

Testing Data

Add the image to test to 'sample_img' Modify 'run.py' to use the new images 'image_name'

python3 run.py

---

Files

gen_anchors.py

prints anchors to use in config

train.py

train the counter

run.py

test using image

tracker.py

test using video (WIP)

---

Configuration

Adding Classes

labels.txt

add class names here, each on a new line

gen_anchors.py

update 'cfg_classes'

run 'python3 gen_anchors.py'

note down details to use next

cfg/yolo_ls.cfg

under [region], adjust 'num' and 'anchors', adjust 'filters' from convolutional layer above

Adjusting YOLO Image Resolution

cfg/yolo_ls.cfg

adjust (at top) 'width' and 'height' of image to multiples of 32

gen_anchors.py

match 'cfg_width' and 'cfg_height'