I am almost done with the. second course of the Machine Learning specialisation on coursera.
I have a set of almost 131000 pictures of 16 cameras’ pictures. I want to create an Ai model that can classify these pictures based on the shooting of the camera. I manually classified around 1300 pics so I can use them to train the model.
do you think that the knowledge I gained so far from those two courses is enough to implement such a model?
if yes, may anyone help me with like having a plan for that model. if no, what kind of other knowledge I should gain before digging deep into it.
Seems like you could simplify the task by masking off all but the corner that has the text, then train a simple classifier to identify the patterns of pixels.
Its very much like the classic “identify handwritten digits” exercise that is found in most courses about neural networks.
and what shall I be doing if I want the model to consider the whole picture not just the corner since most of the pictures having the same camera shot have the same pattern. I guess this will be more of a professional work.
You’re right that I have not seen any reference to Haar Cascading Filters anywhere in DLS. I tried searching for that and found several websites including this one, but none of them seem to be courses you can take. Sorry!
DLS Course 4 Convolutional Networks covers several object recognition and labeling algorithms including YOLO and U-Net for Semantic Segmentation. Both of those are in Week 3 of DLS C4.
Sorry if you found my previous response offensive. I meant it just as a joke, but I guess the lesson here is the humor doesn’t always work when just typing as opposed to in person. I have rewritten it. Sorry again!
Sorry, I forgot to answer that part of your question. It all depends on how much background you have. The DLS series involves quite a bit of notation and understanding how Prof Ng structures data and explains things. Some of the notation is common between MLS and DLS, but definitely not all of it. DLS takes a more advanced approach to some of the topics covered in MLS. It also assumes you have some experience using TensorFlow, but it does introduce quite a bit of that.
You can try just taking DLS Course 4 as a start and see how it goes. If you find it too hard without the previous background, then the alternative plan would be to take DLS Course 1 and Course 2 first. You can skip Course 3 on your first time through, since it is not a prerequisite to Course 4.