Instructions on Training on AUV Underwater Dataset

Clone AlexeyAB Darknet Repo

git clone https://github.com/AlexeyAB/darknet.git

Build Darknet

Modify the makefile to change GPU=0 to GPU=1

Note that this will compile darknet to be training with a GPU, which if you don't have will cause errors. Training on CPU is not efficient and probably not worth attempting, if you want to experiment with CPU training keep GPU=0

In case you get an error regarding nvcc execute

Add your cuda installation directory in path. One way to do that is:

export PATH=/usr/local/cuda-your_cuda_version/bin${PATH:+:${PATH}}

To troubleshoot problems in this stage use

nvidia-smi

You should see CUDA 10.0> and driver version 410>

Importing Images

Add files into the data/images folder

Must be .jpg with .txt file. Format and information can be found on AlexeyAB repostiroy.

Execute python generate_cfg.py

Training

Execute

./train_darketnet.sh

When training on GPU you can get:

CUDA Error: out of memory
CUDA Error: out of memory: File exists

Modify the .cfg file subdivisions to an increased number e.g. 16

Testing

Execute

python yolo.py

For correct class labels modify data/obj.names

Results

python visualize.py