enhancing landmarks extractor by using s3fd second pass inside second pass,
it will be x2 slower, but time will be saved due to more images will be marked properly works on 2GB+
This commit is contained in:
iperov
@@ -3,7 +3,8 @@ import numpy as np
|
||||
import os
|
||||
import cv2
|
||||
from pathlib import Path
|
||||
|
||||
from facelib import FaceType
|
||||
from facelib import LandmarksProcessor
|
||||
|
||||
class LandmarksExtractor(object):
|
||||
def __init__ (self, keras):
|
||||
@@ -23,7 +24,7 @@ class LandmarksExtractor(object):
|
||||
del self.keras_model
|
||||
return False #pass exception between __enter__ and __exit__ to outter level
|
||||
|
||||
def extract_from_bgr (self, input_image, rects):
|
||||
def extract_from_bgr (self, input_image, rects, second_pass_extractor=None):
|
||||
input_image = input_image[:,:,::-1].copy()
|
||||
(h, w, ch) = input_image.shape
|
||||
|
||||
@@ -44,10 +45,29 @@ class LandmarksExtractor(object):
|
||||
landmarks.append ( ( (left, top, right, bottom),pts_img ) )
|
||||
except Exception as e:
|
||||
print ("extract_from_bgr: ", traceback.format_exc() )
|
||||
landmarks.append ( ( (left, top, right, bottom), [ (0,0) for _ in range(68) ] ) )
|
||||
landmarks.append ( ( (left, top, right, bottom), None ) )
|
||||
|
||||
if second_pass_extractor is not None:
|
||||
for i in range(len(landmarks)):
|
||||
rect, lmrks = landmarks[i]
|
||||
if lmrks is None:
|
||||
continue
|
||||
|
||||
image_to_face_mat = LandmarksProcessor.get_transform_mat (lmrks, 256, FaceType.FULL)
|
||||
face_image = cv2.warpAffine(input_image, image_to_face_mat, (256, 256), cv2.INTER_CUBIC)
|
||||
|
||||
rects2 = second_pass_extractor.extract_from_bgr(face_image)
|
||||
if len(rects2) != 1: #dont do second pass if more than 1 face detected in cropped image
|
||||
continue
|
||||
|
||||
rect2 = rects2[0]
|
||||
|
||||
lmrks2 = self.extract_from_bgr (face_image, [rect2] )[0][1]
|
||||
source_lmrks2 = LandmarksProcessor.transform_points (lmrks2, image_to_face_mat, True)
|
||||
landmarks[i] = (rect, source_lmrks2)
|
||||
|
||||
return landmarks
|
||||
|
||||
|
||||
def transform(self, point, center, scale, resolution):
|
||||
pt = np.array ( [point[0], point[1], 1.0] )
|
||||
h = 200.0 * scale
|
||||
|
||||
+21
-12
@@ -52,7 +52,7 @@ class ExtractSubprocessor(Subprocessor):
|
||||
nnlib.import_all (device_config)
|
||||
self.e = facelib.S3FDExtractor()
|
||||
else:
|
||||
raise ValueError ("Wrond detector type.")
|
||||
raise ValueError ("Wrong detector type.")
|
||||
|
||||
if self.e is not None:
|
||||
self.e.__enter__()
|
||||
@@ -61,6 +61,11 @@ class ExtractSubprocessor(Subprocessor):
|
||||
nnlib.import_all (device_config)
|
||||
self.e = facelib.LandmarksExtractor(nnlib.keras)
|
||||
self.e.__enter__()
|
||||
if device_config.gpu_vram_gb[0] >= 2:
|
||||
self.second_pass_e = facelib.S3FDExtractor()
|
||||
self.second_pass_e.__enter__()
|
||||
else:
|
||||
self.second_pass_e = None
|
||||
|
||||
elif self.type == 'final':
|
||||
pass
|
||||
@@ -76,7 +81,7 @@ class ExtractSubprocessor(Subprocessor):
|
||||
|
||||
filename_path_str = str(filename_path)
|
||||
if self.cached_image[0] == filename_path_str:
|
||||
image = self.cached_image[1]
|
||||
image = self.cached_image[1] #cached image for manual extractor
|
||||
else:
|
||||
image = cv2_imread( filename_path_str )
|
||||
|
||||
@@ -102,8 +107,14 @@ class ExtractSubprocessor(Subprocessor):
|
||||
image = image[0:h-hm,0:w-wm,:]
|
||||
self.cached_image = ( filename_path_str, image )
|
||||
|
||||
src_dflimg = None
|
||||
h, w, ch = image.shape
|
||||
if h == w:
|
||||
#extracting from already extracted jpg image?
|
||||
if filename_path.suffix == '.jpg':
|
||||
src_dflimg = DFLJPG.load ( str(filename_path) )
|
||||
|
||||
if self.type == 'rects':
|
||||
h, w, ch = image.shape
|
||||
if min(w,h) < 128:
|
||||
self.log_err ( 'Image is too small %s : [%d, %d]' % ( str(filename_path), w, h ) )
|
||||
rects = []
|
||||
@@ -116,18 +127,13 @@ class ExtractSubprocessor(Subprocessor):
|
||||
rects = data[1]
|
||||
if rects is None:
|
||||
landmarks = None
|
||||
else:
|
||||
landmarks = self.e.extract_from_bgr (image, rects)
|
||||
else:
|
||||
landmarks = self.e.extract_from_bgr (image, rects, self.second_pass_e if src_dflimg is None else None)
|
||||
|
||||
return [str(filename_path), landmarks]
|
||||
|
||||
elif self.type == 'final':
|
||||
src_dflimg = None
|
||||
(h,w,c) = image.shape
|
||||
if h == w:
|
||||
#extracting from already extracted jpg image?
|
||||
if filename_path.suffix == '.jpg':
|
||||
src_dflimg = DFLJPG.load ( str(filename_path) )
|
||||
|
||||
|
||||
result = []
|
||||
faces = data[1]
|
||||
@@ -139,7 +145,10 @@ class ExtractSubprocessor(Subprocessor):
|
||||
face_idx = 0
|
||||
for face in faces:
|
||||
rect = np.array(face[0])
|
||||
image_landmarks = np.array(face[1])
|
||||
image_landmarks = face[1]
|
||||
if image_landmarks is None:
|
||||
continue
|
||||
image_landmarks = np.array(image_landmarks)
|
||||
|
||||
if self.face_type == FaceType.MARK_ONLY:
|
||||
face_image = image
|
||||
|
||||
+6
-1
@@ -64,6 +64,7 @@ class device:
|
||||
|
||||
self.cpu_only = (len(self.gpu_idxs) == 0)
|
||||
|
||||
|
||||
if not self.cpu_only:
|
||||
self.gpu_names = []
|
||||
self.gpu_compute_caps = []
|
||||
@@ -73,7 +74,11 @@ class device:
|
||||
self.gpu_compute_caps += [ device.getDeviceComputeCapability(gpu_idx) ]
|
||||
self.gpu_vram_gb += [ device.getDeviceVRAMTotalGb(gpu_idx) ]
|
||||
self.cpu_only = (len(self.gpu_idxs) == 0)
|
||||
|
||||
else:
|
||||
self.gpu_names = ['CPU']
|
||||
self.gpu_compute_caps = [99]
|
||||
self.gpu_vram_gb = [0]
|
||||
|
||||
if self.cpu_only:
|
||||
self.backend = "tensorflow-cpu"
|
||||
|
||||
|
||||
Reference in New Issue
Block a user