Annoyed by our housemates (we are each other's housemates :/), we want to build an application to friendly 😚 remind each other to clean their dirty dishes put on the kitchen table on time by publicly shaming their behaviour on discord.
What it does
Dish Shamer applies YOLO5 and OpenCV to detect both dirty dishes on the table, and the people who put them, and publicly shame them on their household discord group.
How we built it
We first built our own dataset using a combination of manual photography, manual labelling, and downloading. Then, we realised we didn't actually have enough time left to train the model, so we used the pretrained weights instead. We also built a discord bot and a bunch of application logic to turn the image detection software into an application that can actually detect dirty dishes, recognise faces, and put them together to do some public shaming.
Challenges we ran into
- Train a new YOLO5 model on our own takes far too long time (several hours)
- Get object recognition works nicely
Accomplishments that we're proud of
- Get the software work
- Build the logic of detecting the same items in YOLO frame by frame -> add memory to YOLO, which is quite novel
- Debug within 3h
- Still love each other as housemates so far :D
What we learned
- Apply advanced computer vision into our project
- Fancy debugging skills
- Not sleeping during DurHack is unhealthy :/
What's next for Dish Shamer
We realised sometimes it's sometimes because of the fact that some of our housemates didn't take their clean dished out of the dish washer that other housemates have to leave their dirty dishes on the kitchen table. SO DishWasher Shamer is on the way!