search

shannanyinxiang / SPTS

Official implementation of SPTS: Single-Point Text Spotting (ACM MM 2022 Oral)
137 stars 12 forks source link

Evaluation on Custom dataset #10

Closed k-styles closed 1 year ago

k-styles commented 1 year ago

Hi, I am using your work in my project on a custom dataset for text spotting. I had a doubt on the evaluation folder - I would like to use my custom dataset to train with this model, but I am not really aware on how should i setup the evaluation folder. It would be great if you could let me know about how you have prepared the evaluation files for icdar and totaltext datasets.

shannanyinxiang commented 1 year ago

Taking TotalText for example, the gt_totaltext folder contains .txt annotation files, each for a single test sample. The filename of the .txt annotation file is the image_id (provided in .json annotations) of the corresponding test sample.

The .txt annotation file contains multiple lines, each of which corresponds to a text instance. The format of one line is as follows.

x_1,y_1,x_2,y_2,...,x_n,y_n,####content

where the x_1,y_1,x_2,y_2,...,x_n,y_n are the coordinates of the polygon, and the content is the recognition annotation.