Skip to content

Fix NaN handling in face bbox detection (#13) #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ansorre wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix NaN handling in face bbox detection (#13) #22

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 65 additions & 2 deletions utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,19 @@ def get_face_bboxes(kp2ds, scale, image_shape, ratio_aug):
min_x, min_y = np.min(kp2ds_face, axis=0)
max_x, max_y = np.max(kp2ds_face, axis=0)

# Check for NaN values in face keypoints (happens when face is not visible)
if np.isnan(min_x) or np.isnan(max_x) or np.isnan(min_y) or np.isnan(max_y):
# Return a fallback bbox centered on the image, approximately 30% of image size
fallback_width = int(w * 0.3)
fallback_height = int(h * 0.3)
center_x = w // 2
center_y = h // 2
return [
max(center_x - fallback_width // 2, 0),
min(center_x + fallback_width // 2, w),
max(center_y - fallback_height // 2, 0),
min(center_y + fallback_height // 2, h)
]

initial_width = max_x - min_x
initial_height = max_y - min_y
Expand Down Expand Up @@ -291,10 +304,47 @@ def get_frame_indices(frame_num, video_fps, clip_length, train_fps):

def get_face_bboxes(kp2ds, scale, image_shape):
h, w = image_shape
kp2ds_face = kp2ds.copy()[1:] * (w, h)

# Check for NaN BEFORE any operation
kp2ds_raw = kp2ds.copy()[1:]

# Multiple checks for NaN detection
has_nan = False
try:
has_nan = np.any(np.isnan(kp2ds_raw))
except:
has_nan = True

if has_nan or kp2ds_raw.size == 0:
# Return a fallback bbox centered on the image, approximately 30% of image size
fallback_width = int(w * 0.3)
fallback_height = int(h * 0.3)
center_x = w // 2
center_y = h // 2
return [
max(center_x - fallback_width // 2, 0),
min(center_x + fallback_width // 2, w),
max(center_y - fallback_height // 2, 0),
min(center_y + fallback_height // 2, h)
]

kp2ds_face = kp2ds_raw * (w, h)

min_x, min_y = np.min(kp2ds_face, axis=0)
max_x, max_y = np.max(kp2ds_face, axis=0)

# Additional check after min/max computation
if np.isnan(min_x) or np.isnan(max_x) or np.isnan(min_y) or np.isnan(max_y):
fallback_width = int(w * 0.3)
fallback_height = int(h * 0.3)
center_x = w // 2
center_y = h // 2
return [
max(center_x - fallback_width // 2, 0),
min(center_x + fallback_width // 2, w),
max(center_y - fallback_height // 2, 0),
min(center_y + fallback_height // 2, h)
]

initial_width = max_x - min_x
initial_height = max_y - min_y
Expand All @@ -313,5 +363,18 @@ def get_face_bboxes(kp2ds, scale, image_shape):
expanded_max_x = min(max_x + delta_width, w)
expanded_min_y = max(min_y - 3 * delta_height, 0)
expanded_max_y = min(max_y + delta_height, h)

# Final safety check before converting to int
if np.isnan(expanded_min_x) or np.isnan(expanded_max_x) or np.isnan(expanded_min_y) or np.isnan(expanded_max_y):
fallback_width = int(w * 0.3)
fallback_height = int(h * 0.3)
center_x = w // 2
center_y = h // 2
return [
max(center_x - fallback_width // 2, 0),
min(center_x + fallback_width // 2, w),
max(center_y - fallback_height // 2, 0),
min(center_y + fallback_height // 2, h)
]

return [int(expanded_min_x), int(expanded_max_x), int(expanded_min_y), int(expanded_max_y)]
return [int(expanded_min_x), int(expanded_max_x), int(expanded_min_y), int(expanded_max_y)]