Skip to content

Commit

Permalink
Face Detection Library Change
Browse files Browse the repository at this point in the history
  • Loading branch information
@oguzhanbsolak
oguzhanbsolak committed Jun 25, 2024
1 parent a8938a5 commit 32ef66e
Show file tree
Hide file tree
Showing 5 changed files with 54 additions and 68 deletions.
10 changes: 7 additions & 3 deletions Examples/MAX78000/CNN/facial_recognition/db_gen/generate_face_db.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,12 +21,13 @@
"""
Script to generate Face Id embeddings
"""
import torch
import argparse
import numpy as np
import os
import os.path as path
from ai85.ai85_adapter import AI85SimulatorAdapter
from hawk_eyes.face import RetinaFace
from batch_face import RetinaFace

from utils import append_db_file_from_path, create_weights_include_file, create_embeddings_include_file, create_baseaddr_include_file

Expand All @@ -42,8 +43,11 @@ def create_db_from_folder(args):


ai85_adapter = AI85SimulatorAdapter(MODEL_PATH)
face_detector = RetinaFace(model_name='retina_l', conf=0.1)


if torch.cuda.is_available():
face_detector = RetinaFace(gpu_id=torch.cuda.current_device(), network="resnet50")
else:
face_detector = RetinaFace(gpu_id=-1, network="resnet50")
os.makedirs(args.db, exist_ok=True)

emb_array, recorded_subject = append_db_file_from_path(args.db, face_detector, ai85_adapter)
Expand Down
28 changes: 12 additions & 16 deletions Examples/MAX78000/CNN/facial_recognition/db_gen/utils.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,6 @@
from collections import defaultdict
import numpy as np


from cv2 import imread
from PIL import Image, ExifTags
import torch
Expand Down Expand Up @@ -101,7 +100,6 @@ def append_db_file_from_path(folder_path, face_detector, ai85_adapter):

img = get_face_image(img, face_detector)
if img is not None:
img = ((img+1)*128)
img = (img.squeeze()).detach().cpu().numpy()
img = img.astype(np.uint8)
img = img.transpose([1, 2, 0])
Expand All @@ -118,28 +116,26 @@ def append_db_file_from_path(folder_path, face_detector, ai85_adapter):
emb_id += 1
summary[subject] += 1

#np.save('emb_array.npy', emb_array)
print('Database Summary')
for key in summary:
print(f'\t{key}:', f'{summary[key]} images')
#Format summary for printing image counts per subject



return emb_array, recorded_subject

def get_face_image(img, face_detector):
"""Detects face on the given image
"""
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
bboxes, lndmrks = face_detector.detect(img)
try:
pbox = bboxes[0]
except IndexError:
print('No face detected')
return None

faces = face_detector(img)
box, landmarks, score = faces[0]

pbox = box
for i in range(4):
pbox[i] = np.clip(pbox[i], 0, None)

img = torch.Tensor(img.transpose([2, 0, 1])).to(device).unsqueeze(0)

img = Normalize_Img(img) #Normalize image for faceID

# Convert bounding box to square
height = pbox[3] - pbox[1]
Expand All @@ -153,7 +149,7 @@ def get_face_image(img, face_detector):
# Crop image with the square bounding box
img = VF.crop(img=img, top=int(pbox[1]), left=int(pbox[0]),
height=int(pbox[3]-pbox[1]), width=int(pbox[2]-pbox[0]))

# Check if the cropped image is square, if not, pad it

_, _, h, w = img.shape
Expand Down Expand Up @@ -224,7 +220,7 @@ def create_weights_include_file(emb_array, weights_h_path, baseaddr):
h_file.write('#define KERNELS_3 { \\\n ')
for dim in range(Embedding_dimension):
init_proccessor = False
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes TODO: Change this from Hardcoded
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes
reindex = i + 8 - 2*(i%9)
if reindex < 1024: # Total emb count is 1024, last index 1023
single_byte = str(format(emb_array[reindex][dim], 'x')) #Relocate emb for cnn kernel
Expand Down Expand Up @@ -265,7 +261,7 @@ def create_weights_include_file(emb_array, weights_h_path, baseaddr):
four_byte = 0
data_arr = bytearray(np.uint8([data_arr]))
for dim in range(Embedding_dimension):
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes TODO: Change this from Hardcoded
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes
reindex = i + 8 - 2*(i%9)
if reindex < 1024: # Total emb count is 1024, last index 1023
single_byte = int(emb_array[reindex][dim]) #Relocate emb for cnn kernel
Expand Down
10 changes: 7 additions & 3 deletions Examples/MAX78002/CNN/facial_recognition/db_gen/generate_face_db.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,12 +21,13 @@
"""
Script to generate Face Id embeddings
"""
import torch
import argparse
import numpy as np
import os
import os.path as path
from ai85.ai85_adapter import AI85SimulatorAdapter
from hawk_eyes.face import RetinaFace
from batch_face import RetinaFace

from utils import append_db_file_from_path, create_weights_include_file, create_embeddings_include_file, create_baseaddr_include_file

Expand All @@ -41,8 +42,11 @@ def create_db_from_folder(args):


ai85_adapter = AI85SimulatorAdapter(MODEL_PATH)
face_detector = RetinaFace(model_name='retina_l', conf=0.1)


if torch.cuda.is_available():
face_detector = RetinaFace(gpu_id=torch.cuda.current_device(), network="resnet50")
else:
face_detector = RetinaFace(gpu_id=-1, network="resnet50")
os.makedirs(args.db, exist_ok=True)

emb_array, recorded_subject = append_db_file_from_path(args.db, face_detector, ai85_adapter)
Expand Down
72 changes: 27 additions & 45 deletions Examples/MAX78002/CNN/facial_recognition/db_gen/utils.py
View file
Original file line number Diff line number Diff line change
@@ -1,35 +1,22 @@
################################################################################
# Copyright (C) 2022 Maxim Integrated Products, Inc., All Rights Reserved.
###############################################################################
#
# Permission is hereby granted, free of charge, to any person obtaining a
# copy of this software and associated documentation files (the "Software"),
# to deal in the Software without restriction, including without limitation
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
# and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
# Copyright (C) 2022-2023 Maxim Integrated Products, Inc. (now owned by
# Analog Devices, Inc.),
# Copyright (C) 2023-2024 Analog Devices, Inc.
#
# The above copyright notice and this permission notice shall be included
# in all copies or substantial portions of the Software.
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
# OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
# IN NO EVENT SHALL MAXIM INTEGRATED BE LIABLE FOR ANY CLAIM, DAMAGES
# OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
# ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
# OTHER DEALINGS IN THE SOFTWARE.
# http://www.apache.org/licenses/LICENSE-2.0
#
# Except as contained in this notice, the name of Maxim Integrated
# Products, Inc. shall not be used except as stated in the Maxim Integrated
# Products, Inc. Branding Policy.
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# The mere transfer of this software does not imply any licenses
# of trade secrets, proprietary technology, copyrights, patents,
# trademarks, maskwork rights, or any other form of intellectual
# property whatsoever. Maxim Integrated Products, Inc. retains all
# ownership rights.
#
###############################################################################
##############################################################################

"""
Utility functions to generate embeddings and I/O operations
Expand All @@ -43,7 +30,6 @@
from cv2 import imread
from PIL import Image, ExifTags
import torch
import torchvision
import torchvision.transforms.functional as VF


Expand Down Expand Up @@ -115,7 +101,6 @@ def append_db_file_from_path(folder_path, face_detector, ai85_adapter):

img = get_face_image(img, face_detector)
if img is not None:
img = ((img+1)*128)
img = (img.squeeze()).detach().cpu().numpy()
img = img.astype(np.uint8)
img = img.transpose([1, 2, 0])
Expand All @@ -133,29 +118,26 @@ def append_db_file_from_path(folder_path, face_detector, ai85_adapter):
emb_id += 1
summary[subject] += 1

#np.save('emb_array.npy', emb_array)
print('Database Summary')
for key in summary:
print(f'\t{key}:', f'{summary[key]} images')
#Format summary for printing image counts per subject



return emb_array, recorded_subject

def get_face_image(img, face_detector):
"""Detects face on the given image
"""
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
bboxes, lndmrks = face_detector.detect(img)
try:
pbox = bboxes[0]
except IndexError:
print('No face detected')
return None
img = torch.Tensor(img.transpose([2, 0, 1])).to(device).unsqueeze(0)

img = Normalize_Img(img) #Normalize image for faceID

faces = face_detector(img)
box, landmarks, score = faces[0]

pbox = box
for i in range(4):
pbox[i] = np.clip(pbox[i], 0, None)

img = torch.Tensor(img.transpose([2, 0, 1])).to(device).unsqueeze(0)

# Convert bounding box to square
height = pbox[3] - pbox[1]
width = pbox[2] - pbox[0]
Expand All @@ -168,7 +150,7 @@ def get_face_image(img, face_detector):
# Crop image with the square bounding box
img = VF.crop(img=img, top=int(pbox[1]), left=int(pbox[0]),
height=int(pbox[3]-pbox[1]), width=int(pbox[2]-pbox[0]))

# Check if the cropped image is square, if not, pad it

_, _, h, w = img.shape
Expand Down Expand Up @@ -238,7 +220,7 @@ def create_weights_include_file(emb_array, weights_h_path, baseaddr):
h_file.write('#define KERNELS_3 { \\\n ')
for dim in range(Embedding_dimension):
init_proccessor = False
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes TODO: Change this from Hardcoded
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes
reindex = i + 8 - 2*(i%9)
if reindex < 1024: # Total emb count is 1024, last index 1023
single_byte = str(format(emb_array[reindex][dim], 'x')) #Relocate emb for cnn kernel
Expand Down Expand Up @@ -279,7 +261,7 @@ def create_weights_include_file(emb_array, weights_h_path, baseaddr):
four_byte = 0
data_arr = bytearray(np.uint8([data_arr]))
for dim in range(Embedding_dimension):
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes TODO: Change this from Hardcoded
for i in range(emb_array.shape[0] + 4): # nearest %9 == 0 for 1024 is 1027, it can be kept in 1028 bytes
reindex = i + 8 - 2*(i%9)
if reindex < 1024: # Total emb count is 1024, last index 1023
single_byte = int(emb_array[reindex][dim]) #Relocate emb for cnn kernel
Expand Down
2 changes: 1 addition & 1 deletion Examples/MAX78002/CNN/facial_recognition/src/record.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -655,7 +655,7 @@ int add_person(Person *p)
if (face_detected) {
printf("Box width: %d\n", box[2] - box[0]);
printf("Box height: %d\n", box[3] - box[1]);
if ((box[2] - box[0]) < 90 || (box[3] - box[1]) < 130) {
if ((box[2] - box[0]) < 70 || (box[3] - box[1]) < 110) {
face_detected = 0;

if (!init_come_closer) {
Expand Down

0 comments on commit 32ef66e

Please sign in to comment.