yolov3-aichamp-workshop

This guide assumes that you are using windows OS.

Step 0:

Before you start any of the following steps, it's smart to start downloading files to instantiate the YOLO network.

Note: We have provided a utility function which download and places files where you want them. Introduction.ipynb has a step which will download the files we need. However, as the weights file is quite big, it can be smart to start it now.

For introduction.ipynb, we need the following files:

YOLOv3 configuration file

YOLOv3 weight file

COCO labels file

We have added a couple more configuration and their associated weights in the table below (with some extra stats for the nerds among us ;)):

config	weight	FLOPs	FPS	mAP
config	weight	141.45 Bn	20	60.6
config	weight	5.56 Bn	220	33.1

Note If you decide to download your own config, make sure you download the raw version of the file as shown in the image below:

More options can be found here. Keep in mind that this implementation might not support everything there.

Note that YOLO-320, YOLO-416 and YOLO-608 actually use the same configuration and weights. The difference is in the way we tell YOLO to scale the input. We will get back to this later in the workshop. It is infact the configuration and weight you were told to download at the start, so no need to download them again!

Step 1:

Download and install python 3.X (This workshop was made using 3.7.5. As of making this workshop, there is a bug with 3.8, so don't use it).

Verify install with py -V using CMD.

Step 2:

Download this repo to your computer: git clone https://github.com/Adsime/yolov3-aichamp-workshop.git.

cd into the project.

Step 3:

It's recommended that you create a virtual environment. If you don't want to, skip to step 4.

Create a virtual environment:

py -m venv ./venv

Activate it:

.\venv\Scripts\activate.bat

If your terminal looks something like this, then the virtual environment is activated:

Step 4:

Install dependacies:

pip install -r requirements.txt

Step 5:

Start jupyter notebook:

jupyter notebook

There are 3 notebooks made for this workshop.

• introduction.ipynb will give an overview of all the code in the project.
• image_workshop.ipynb will do the same as introduction, but let you edit the parameters more easily yourself to play around with.
• video_workshop.ipynb uses the youtube_dl library, which lets us download youtube videos and perform live object detection on them.