1
$\begingroup$

I am given a selection of videos of users exploring simulated 3D enviroments (kind of looks like the Sims video game) and I am tasked with being able to classify each room using a tensorflow framework. E.g. if a user were to hover over a certain area in the environment, the model should be able to classify whether it's a security line

enter image description here

or baggage check etc.

enter image description here

I am fairly new to machine learning algorithms and techniques, however I have been given a working VGG16 CNN model that is able to characterize (given labeled input data) actors in a video clip, and I am expected to rework this into environment classification. Is such a thing possible given this framework? I am undecided on whether I should collect images from various angles of the room/environment, and labeling and training each image as a whole using VGG16/Resnet50, or to implement some sort of R-CNN/YOLO model using bounding boxes to classify the environment given known objects found on screen. I am not actually limited to the given tensorflow framework, so if there is a better way to approach this problem I can recreate a new code environment from scratch if needed.

$\endgroup$

1 Answer 1

0
$\begingroup$

It depends on the objective you have: if the objective is just to recognize the environment, choose a simple image recognition model first. Furthermore, simulated environment are cleaner than real ones (no shadows, no noise). Consequently, there are good chances to reach very good results with a fast model (like Inception) that would also have a very good processing time.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.