Scene recognition using object detection.
1. Caffe must not be used and the OS should be windows 10.
2. Coding help can be taken from online sources.
3. Using the images and their annotations (dataset will be provided by me), train the capsule neural network, Fast-RCNN, Faster RCNN and Mask RCNN.
4. Record validation accuracy for each trained neural network.
5. Using the testing images (also provided by me) detect the objects in the scene.
6. On the basis of objects detected in the scene, predict the scene. Example of the same can be derived from the images shown below. However, in this case there will be more objects to detect in a scene and on that basis the scene has to be recognized.
![1530_figure.jpg](https://secure.tutorsglobe.com/CMSImages/1530_figure.jpg)
7. Each image must be tested using all the trained neural networks.
8. Moreover, if none of the scene according to object detection receives above 50% accuracy then the system must show a message ‘don't know'.
9. All the images tested must be provided in jpeg format in a separate folder with name results.
10. All the codes must be provided in .py format and the code must contain comments that explain each part of the code.
Attachment:- dataset.rar