how to train custom dataset using deep sortmodel?

[IamExperimenting]

@abhyantrika , hope you are doing great. I have a question on "how to train the deep sort model using custom dataset" . Let's say, I have 3 classes namely car, motorbike, bus.
I have a doubt about creating the folders here.

|__car
|    |__car_image1.jpg
|    |__car_image2.jpg
|    |__car_image3.jpg
|__bus
|    |__bus_image1.jpg
|    |__bus_image2.jpg
|    |__bus_image3.jpg
|__motorbike
|    |__motorbike_image1.jpg
|    |__motorbike_image2.jpg
|    |__motorbike_image3.jpg

I have another doubt when I downloaded the example from this repository under the crops folder there are 329 folders and each folder contains samples of the same car images at a different angle. Here my questions raise, do we need to create a folder for each and every car like a separate folder green colour car and place the images of the green colour car inside it. And a separate folder for red colour car and place the images of the red colour car inside it. should it be like the below folder structure?

|__car
|    |_green_car
|    |  |__green_car_image1.jpg
|    |  |__green_car_image2.jpg
|    |  |__green_image3.jpg
|    |_red_car
|    |  |__red_car_image1.jpg
|    |  |__red_car_image2.jpg
|    |  |__red_car_image3.jpg

Last question, what would be the minimum number of images to be considered for each class?

[abhyantrika]

The objective is to track the same car across different camera angles.
Hence, we would like our model to learn features which are invariant to different views.

[IamExperimenting]

@abhyantrika thanks for your response. Just to clarify. Let's say the model needs to learn the green colour Toyota car in all different views. correct me if I'm wrong.

Could you please also provide us with the dataset folder structure?