segmentation.md

Semantic Segmentation

The model that I used to implement Semantic Segmentation was COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x. Using detectron2, I was able to implement the model and train it on the COCO dataset. I did not adjust any of the parameters of the model, but I did adjust the parameters of the training. I trained the model for 2000 iterations, and I used a learning rate of 0.00025. I also used a batch size of 2. I used the default parameters for the rest of the training. Watching the final resulting video, I can see that it did a fabolous job at segmenting the objects in the video. However it was not completely accurate. It did not segment the objects perfectly, but it did a good job at it. I think that if I had trained the model for more iterations, it would have been more accurate. I also think that if I had used a larger batch size, it would have been more accurate. Overall, I think that the model did a good job at segmenting the objects in the video, but it could have been better.