|
|
|
|
|
|
|
|
|
|
|
import ast
|
|
|
import pandas as pd
|
|
|
import os
|
|
|
import math
|
|
|
import random
|
|
|
random.seed(0)
|
|
|
|
|
|
|
|
|
dir_path = "visual_discrimination/sweep/geometric_dataset"
|
|
|
df = pd.read_csv(os.path.join(dir_path, "dataset_dump.csv"))
|
|
|
|
|
|
'''
|
|
|
The dataset_dump.csv file contains the following columns:
|
|
|
1. filename: SVG filename of the image
|
|
|
2. shape_dictionary: It contains shape name as the key and corresponding count as the value
|
|
|
'''
|
|
|
|
|
|
data = []
|
|
|
for index, row in df.iterrows():
|
|
|
shape_dict = ast.literal_eval(row['shape_dictionary'])
|
|
|
filename = row['filename']
|
|
|
sweep = ast.literal_eval(row['sweep'])
|
|
|
innerlist = []
|
|
|
for shape, count in shape_dict.items():
|
|
|
question = f"Count the total number of {shape}s in the image, including each concentric {shape} separately. For example, if there is one {shape} with 2 inner concentric rings, that counts as 3 {shape}s. Respond with only a number."
|
|
|
answer = count
|
|
|
innerlist.append({
|
|
|
'filename': filename,
|
|
|
'question': question,
|
|
|
'answer': answer,
|
|
|
'sweep': sweep
|
|
|
})
|
|
|
innerlist = random.sample(innerlist, 1)
|
|
|
data.extend(innerlist)
|
|
|
|
|
|
df = pd.DataFrame(data)
|
|
|
df.to_csv(os.path.join(dir_path, "dataset_info.csv"), index=False)
|
|
|
|
