- Implement a computer vision algorithm that is able to detect logos in the flickr-27 dataset
- What would you do if you had to detect a new unlabelled logo?
- Why the solution proposed in https://arxiv.org/pdf/1803.11417.pdf is adapted to "real-world dynamic applications"?
- Dataset location: http://image.ntua.gr/iva/datasets/flickr_logos/
- Provide a quick explanation of the proposed approach and the motivation behind it
- Use Keras, Torch, or Tensorflow
- You can use any pre-trained model
- Inference time performance must not be left out
- Fork the project via
github - Clone your forked repository project https://github.com/YOUR_USERNAME/synthesio-vision-test
- Commit and push your different modifications
- Don't forget to explain how to run your code and/or get your results
- Explain your choices, why you did things this way (in a file or in comments or the way you want)