Rabbit Model 1 Computer Vision Project

This model is an image segmentation classifier that can identify between the two breeds of rabbit native to Maryland – The Eastern Cottontail (Sylvilagus floridanus) and the Appalachian Cottontail (Sylvilagus obscurus), and a third category encompassing all domestic rabbits (Oryctolagus cuniculus domesticus). The reason for this would be around Easter especially each year, many people adopt rabbits, without realizing how difficult it is to care for them. They then will dump the domestic rabbits on the street, instead of a shelter, falsely believing that they have the survival skills of native species. Unfortunately these rabbits will almost certainly die. Hopefully with the help of the model, people will take more action in the death prevention of some beloved pets.

In addition to this, the model has other use cases, such as population control, as despite being maladapted for the enviornment, a group of feral domestic rabbits can spread disease. The app can also be used for population control numbers, as a tool to help prevent the mislabeling of species that look very similar. Additionally, it could be used to monitor population counts in various areas of Maryland of all three species. This makes the intended audience of the model any animal enthusiest or health professional, as it could be interesting to both groups.

I opted for image segmentation on this model due to the ear size difference being a key identifying metric between the Cottontails. It currently has 3 classes, one for Appalachian Cottontail (Sylvilagus obscurus), one for the Eastern Cottontail (Sylvilagus floridanus), and last for domestic rabbits (Oryctolagus cuniculus domesticus). In the next version of the model I will start filling out the null class in order to avoid false positives.

The sample size per class is quite evenly distributed. The domestic rabbits (Oryctolagus cuniculus domesticus) have the most sampled images with 77. This was due to the fact that there is a large variety of colors and coats for domestic rabbits (Oryctolagus cuniculus domesticus) and I needed to sample a wide range of them to compensate for this. I have broken down the model into the default 70 - 20 - 10 distribution, although I plan to experiment with the distribution in further iterations. The next highest was the The Eastern Cottontail (Sylvilagus floridanus). They had 60 images attributed to them. Lastly, I sampled only 54 Appalachian Cottontail (Sylvilagus obscurus) due to the fact that they are the rarest species of the three, and the least likely for people to come across. In the future, I plan to increase the amount of rabbits per class to further improve the training. I pulled images only from iNaturalist for all the dataset, specifically the images with the approved “Research Grade '' tag. I made sure to use a wide range of ages of each group of rabbits, including images of deceased rabbits as well. This is due to the fact that I am unsure of what state someone would want to identify the rabbit. These species do not look much different in the winter, although the domestic rabbits tend to have more trouble with clotted shedding after an extended period of time in the wild. This means that older domestic rabbits tend to have lost the smooth appearance of their wild counterparts. I have used exclusively outdoor images of domestic rabbits so that the model does not associate the indoors with the domestic breed.

This model was created for a class assignment in AI and Natural History at St. Mary’s College of Maryland.