[IMPORTANT] keypoint issue in merged_25pts_ldmk.txt

[eungbean]

The SHHQ-1.0 dataset has a size of 512 x 1024.
However, some keypoints have x-values that exceed 512.
For example,

For example,

image_000023.jpg 595 150 517 211 478 211 417 345 473 456 550 211 -1 -1 -1 -1 512 467 478 467 461 695 434 918 545 467 500 695 478 896 578 133 601 133 528 122 -1 -1 556 940 556 929 473 912 500 957 478 962 417 934

... These keypoints produce the following results.

( in 18-keypoint visualization)

I inspected the merged_25pts_ldmk.txt file using the following code.

ldmk_path = osp.join(self.root_dir, "merged_25pts_ldmk.txt")
with open(ldmk_path, "r") as f:
    lines = f.readlines()

error_hits = 0
for line in lines:
    line = line.strip().split(" ")
    key = line[0]
    keypoints = np.array(line[1:], dtype=np.float32).reshape(-1, 2)
    
    # if x is larger than 512, increase error hits
    if np.any(keypoints[:, 0] > 512): 
       error_hits += 1

print(error_hits)

After conducting the inspection using the above code, it was found that there are a total of 1917 keypoints with such errors.

To address this issue, upon further investigation, it was discovered that the keypoints were based on a 1024 x 1024 reference.

As a solution, subtracting 256 from the x-values yielded the following results.

    if np.any(keypoints[:, 0] > 512): 
       keypoints -= np.array([256,0])

I attatched code for fix bug in merged_25pts_ldmk.txt.
just run the code through txt file.

import os.path as osp
import numpy as np


ldmk_path = osp.join("merged_25pts_ldmk.txt")
with open(ldmk_path, "r") as f:
    lines = f.readlines()

new_lines = []

for line in lines:
    line = line.strip().split(" ")
    key = line[0]
    keypoints = np.array(line[1:], dtype=np.int8).reshape(-1, 2)
    
    # if x is larger than 512, increase error hits
    if np.any(keypoints[:, 0] > 512): 
       keypoints -= np.array([256, 0])
    
    new_line = ' '.join([key] + [str(i) for i in keypoints.reshape(-1)])
    new_lines.append(new_line)
    
with open("new_merged_18pts_ldmk.txt", "w") as f:    
    f.write("\n".join(new_lines))   
    

[eungbean]

fix error.

import os.path as osp
import numpy as np


ldmk_path = osp.join("merged_25pts_ldmk.txt")
with open(ldmk_path, "r") as f:
    lines = f.readlines()

new_lines = []

for line in lines:
    line = line.strip().split(" ")
    key = line[0]
    keypoints = np.array(line[1:], dtype=np.float32).reshape(-1, 2)

    # if x is larger than 512, increase error hits
    if np.any(keypoints[:, 0] > 512):
        missing_keypoint_index = keypoints == -1
        keypoints -= np.array([256, 0])
        keypoints[missing_keypoint_index] = -1

    new_line = " ".join([key] + [str(int(i)) for i in keypoints.reshape(-1)])
    new_lines.append(new_line)

with open("new_merged_25pts_ldmk.txt", "w") as f:
    f.write("\n".join(new_lines))