I have experience with ML and DL (competitions on Kaggle), however I never tried to use YOLO.
I am just doing some research and looks like in torchvision.models there are no models which can be used with YOLO, am I correct?
I found some articles/tutorials how to implement YOLO in PyTorch but they contain training. Does it mean there is no way to divide image into rectangles with objects (like cars, people, etc) without having data to train first?
I want to build system to recognize specific objects on the images from camera. I want to use data which contains only 0/1 labels, so no bounding boxes.
My idea was to divide full images from camera into smaller images and use only these smaller images for training, but for that I need pretrained YOLO model.
YOLO contains the training stage to identify certain objects. I bet there are some pretrained YOLO models out there. You can use them as long as the specific objects you are interested in are the same as the pretrained YOLO model. What is your end goal, do you expect the model to output “YES/NO” on certain objects from camera feed? BTW, if the camera is on iPhone, feel free to check out Apple’s native support. I think they already have a framework to help you detect basic objects, such as a computer, from camera feed: https://developer.apple.com/documentation/vision. Maybe Google has a similar framework on Android.
Also, if your training data is “YES/NO” answer on whether or not certain objects exist in camera feed, and if your trained model only need to output “YES/NO”. I don’t think you have to use YOLO. You can just have a VGG, or AlexNet, but the output can be a 1D vector, where each value represent the probability of certain object exist or not. I high recommend checking out data argument tricks (such as training data rotation, zoom in/out) for this kind of training task.
My idea was to automatically split full size Images from camera and then to manually label cropped Images.
Then during training to split images again and train second model on cropped Images and labels.
Then during test to split new Images with pretrained YOLO model and predict cropped Images with trained model.
My question was about PyTorch models to use, is there any pretrained model I could use for such task or it is not possible.
As I said I am quite experienced with training DL (you can check my Kaggle profile) what I dont know yet is YOLO simply because I never had chance to use such data.