Questions

1. Use pre-trained Fast-RCNN and MobileNet models for real-time object detection with Keras.
2. We used a YOLO v2 pre-trained model to implement object detection. Try to use a YOLO v3 pre-trained model to implement object detection.
3. What is fine-tuning, and how does it differ from transfer learning? Demonstrate with an example.
4. We have trained the FC layers of VGG16 only for transfer learning. Use VGG19, Resnet50, and Inception V3 models instead, with Keras. Does the accuracy improve?
5. For transfer learning with Keras, we used 500 images for training and 50 images for validation for each of the flower classes, with a non-standard 91:9 training to validation dataset ratio. Change it to standard 80:20 validation—how much does it affect the accuracy in the validation dataset?
6. Follow this link, https://cs.stanford.edu/people/jcjohns/papers/eccv16/JohnsonECCV16.pdf, and implement the NST algorithm to transfer the style of image (except Starry Night) to an input content image using transfer learning...