For this week, I tried to make up the works from the last week when I was really busy with the career fairs and interviews. First, I worked on design presentation slides. Second, I downloaded a Product 10K images dataset from JD.com, which contains about 22.5GB size of images. It has a total of 197,310 images of products in JD.com, and it was splitted into test and train sets. There were 55,377 images in the test set and 141,933 images in the train set. I initially had some confusion that I won’t be able to use this dataset due to the Chinese character on the products, but now that I am clarified with my confusion so I started classifying the images that are relevant to our project. As a result, 23,230 images were relevant to our project for the train set, and other irrelevant images were mostly images of clothing.
Lastly, I also found some example codes from Kaggle that ran CNN on the retail product dataset and fruit dataset, and trying to adapt the code into our dataset. These example code used CNN with ResNet.
https://www.kaggle.com/code/lifefilm/10k-resnet50-fastai
https://www.kaggle.com/code/aninditapani/cnn-from-scratch-with-98-accuracy
I am planning to do some more research on how to implementing OCR into our case. Our ideal solution would be performing CNN first and have OCR performed later for some special cases.
For this week, my initial tasks were preparing dataset and testing CNN and OCR until October 2nd. I finished preparing dataset, and still working on CNN and OCR so I am in a risk of falling behind, but I think I would be able to finish the rest of the task by October 2nd.
For next week, I will start working on design document, which is due on October 14th. Also, I am hoping to start implementing and training the model with the dataset I have.