This repository is about the project during DEEP LEARNING CAMP JEJU ( 2018. 7. 1 ~ 30 )
Motivation of Project
Jeju Island POI(point-of-interst) Dataset
Sometime it really hard to find proper place to go.
What if, the Model can understand where do i want to go and understand location of the place in the map ?
The Model will be able to recommend a suitable place according to the Location you are currently in and Keyword you want to go.
We using Relation Networks basically contains CNN for Image processing, and RN for Relational reasoning between two object(Image, Text).
And also using Wild Relation Network for improve the recommendation accuracy.
The data contains the following information
- User Location
- Image Location (Center, Boundary)
- Place Information (type [Restaurant, Cafe, ...], Location [Latitude, Longitude])
At the beginning of the project, I decided to only consider the POI for restaurant and cafe in Jeju Island. Each Image has a number of POIs which marked with red(restaurant), green(cafe) marker.
Also, the center(user position) location of the image, boundary location information and the location information of restaurant and cafe in Jeju Island are required for input.
Among the various Keyword, I will focus on the following Keyword first.
- Restaurant ?
- Cafe ?
- Beachfront ?
- Near ?
- Cafe nearby restaurant ?
- Restaurant nearby cafe ?
The possible Answer is a softmax vector whose with probability for POIs
[probability of POI 1, probability of POI 2, probability of POI 3,...,probability of POI n]
2 - Class Dataset
|RN model||Baseline model|
10 - Class Dataset
|WReN model||RN model||Baseline model|
|Top - 1 Accuracy||33%||20%||0%|
|Top - 3 Accuracy||76%||70%||70%|
- A Simple Neural Network Module for Relational Reasoning by Santoro, Adam, et al.
- Measuring Abstract Reasoning in Neural Networks by Santoro, Adam, et al.
- Relation-Network-Tensorflow by Shao-Hua Sun
This was supported by Deep Learning Camp Jeju 2018 which was organized by TensorFlow Korea User Group. Thank you to my mentor SOURABH BAJAJ who gave me the best mentoring and thank you everyone, that I met during the camp.
© [Sein Jang], 2018. Licensed under the MIT License.