[WEEK 4-Country Classification Using House Photos]
Team Members: Meltem Tokgöz , Enes Furkan Çiğdem , Asma Aiouez
This week, we investigated some papers which ove related to our project context.Image classification, predicting location at on photograph are some of the subjects of the papers.We have found that many researches have been done on image classification problems. We were specifically interested in the ones that used deep learning techniques as that is going to be the method that we’ll be using to train our model.
Some of these articles are as follows:
Applications of image classification fall into both generic models[1-2–3–4] and specific ones, for instance medical images [11], license plate and vehicle recognition [5], recognition of the urban environment [6], fruit recognition [7]. In addition, for the ones that dealt with image classification in the architectural style field and used techniques such as pattern detection [8], support vector machine [9], and [10] which broke images of buildings into “blocklets” to recognize feature vectors to classify with a support vector machine.
[1] https://ieeexplore.ieee.org/document/8404530/
[2] http://islab.ulsan.ac.kr/files/announcement/513/rcnn_pami.pdf
[3] https://people.eecs.berkeley.edu/~jonlong/long_shelhamer_fcn.pdf
[5] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4801602/
[6] https://arxiv.org/ftp/arxiv/papers/1608/1608.03396.pdf
[7] https://www.mdpi.com/1424-8220/16/8/1222/htm
[8] https://www.cs.ccu.edu.tw/~wtchu/projects/VP/ICMR2012-chu.pdf
[9] https://pdfs.semanticscholar.org/1a1c/4e75c74b715fcc0903a044c4f7aa3d3bbf1c.pdf
[10] https://pdfs.semanticscholar.org/0ed0/eb02de7579c714236c480f06faf239f3cd95.pdf
[11] https://arxiv.org/pdf/1706.00712.pdf
These are some of the researches we have mentioned above.
Also this week we tested our datas in a simple single layer neural network code because we collected it and we should trust the dataset before constructing an architecture.
The results are as follows:
Due to the small number of our dataset image, we think our results are low.In addition, the size of the pictures may be effective.We choosen small size in dataset.In short, this week we searched the used methods for our study and we made our first trying.
