Model Training using a Data Set with CSV files and Images

Question

. 。 Answers to this question are eligible for a +50 reputation bounty. Bounty grace period ends in 23 hours . Lakshan Costa wants to to this question: 这个问题：

I'm looking for a method to do this, my guess was it can be done like image captioning. If possible direct me to a code that was done like this. Any help would be appreciated

I'm trying out to retrieve data from vertical bar charts as x , y , x_axis_title , y_axis_title . An example image is as follows.

I am currently creating a dataset with vertical bar charts as above. with a CSV file for each image as follows.

1. Would it be possible to create a model that can output the titles, and values for both x, and y without using OCR(Optical character recognition) with a certain number of images?

2. Is there a specific method to build the model? or a better method?

**I was trying to do this using tesseract OCR but it was a bit inaccurate

Any help would be appreciated!

Answer 1

In my personal experience, Paddle OCR works a lot better than Tesseract and can help you identify all the fields. Here is a good article explaining the usage of Paddle. Using this OCR, you will not really face any problem with the text since most of it is clearly visible

I don't think if it will be possible to get the data without that. As for the values of x and y, you can use Open CV and get intersection points of blue and white to get the points and link them with the y axis to get the exact values