ElenaRyumina
commited on
Commit
β’
c83f8fa
1
Parent(s):
cf60969
Summary
Browse files- .gitignore +2 -1
- app.css +12 -0
- app.py +14 -7
- app/app_utils.py +54 -19
- app/face_utils.py +6 -5
- app/model.py +13 -5
- app/model_architectures.py +150 -0
- config.toml +4 -4
- result.mp4 +0 -0
.gitignore
CHANGED
@@ -168,4 +168,5 @@ dmypy.json
|
|
168 |
.pyre/
|
169 |
|
170 |
# Custom
|
171 |
-
*.pth
|
|
|
|
168 |
.pyre/
|
169 |
|
170 |
# Custom
|
171 |
+
*.pth
|
172 |
+
*.pt
|
app.css
CHANGED
@@ -20,6 +20,14 @@ div.dl2 img {
|
|
20 |
max-height: 200px;
|
21 |
}
|
22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
div.video1 div.video-container {
|
24 |
height: 500px;
|
25 |
}
|
@@ -32,6 +40,10 @@ div.video3 {
|
|
32 |
height: 200px;
|
33 |
}
|
34 |
|
|
|
|
|
|
|
|
|
35 |
div.stat {
|
36 |
height: 286px;
|
37 |
}
|
|
|
20 |
max-height: 200px;
|
21 |
}
|
22 |
|
23 |
+
div.dl5 {
|
24 |
+
max-height: 200px;
|
25 |
+
}
|
26 |
+
|
27 |
+
div.dl5 img {
|
28 |
+
max-height: 200px;
|
29 |
+
}
|
30 |
+
|
31 |
div.video1 div.video-container {
|
32 |
height: 500px;
|
33 |
}
|
|
|
40 |
height: 200px;
|
41 |
}
|
42 |
|
43 |
+
div.video4 {
|
44 |
+
height: 200px;
|
45 |
+
}
|
46 |
+
|
47 |
div.stat {
|
48 |
height: 286px;
|
49 |
}
|
app.py
CHANGED
@@ -18,6 +18,7 @@ from app.app_utils import preprocess_image_and_predict, preprocess_video_and_pre
|
|
18 |
def clear_static_info():
|
19 |
return (
|
20 |
gr.Image(value=None, type="pil"),
|
|
|
21 |
gr.Image(value=None, scale=1, elem_classes="dl2"),
|
22 |
gr.Label(value=None, num_top_classes=3, scale=1, elem_classes="dl3"),
|
23 |
)
|
@@ -27,6 +28,7 @@ def clear_dynamic_info():
|
|
27 |
gr.Video(value=None),
|
28 |
gr.Video(value=None),
|
29 |
gr.Video(value=None),
|
|
|
30 |
gr.Plot(value=None),
|
31 |
)
|
32 |
|
@@ -45,8 +47,9 @@ with gr.Blocks(css="app.css") as demo:
|
|
45 |
)
|
46 |
with gr.Column(scale=2, elem_classes="dl4"):
|
47 |
with gr.Row():
|
48 |
-
output_video = gr.Video(label="Original video", scale=
|
49 |
output_face = gr.Video(label="Pre-processed video", scale=1, elem_classes="video3")
|
|
|
50 |
output_statistics = gr.Plot(label="Statistics of emotions", elem_classes="stat")
|
51 |
gr.Examples(
|
52 |
["videos/video1.mp4",
|
@@ -58,7 +61,7 @@ with gr.Blocks(css="app.css") as demo:
|
|
58 |
gr.Markdown(value=DESCRIPTION_STATIC)
|
59 |
with gr.Row():
|
60 |
with gr.Column(scale=2, elem_classes="dl1"):
|
61 |
-
input_image = gr.Image(type="pil")
|
62 |
with gr.Row():
|
63 |
clear_btn = gr.Button(
|
64 |
value="Clear", interactive=True, scale=1, elem_classes="clear"
|
@@ -67,7 +70,9 @@ with gr.Blocks(css="app.css") as demo:
|
|
67 |
value="Submit", interactive=True, scale=1, elem_classes="submit"
|
68 |
)
|
69 |
with gr.Column(scale=1, elem_classes="dl4"):
|
70 |
-
|
|
|
|
|
71 |
output_label = gr.Label(num_top_classes=3, scale=1, elem_classes="dl3")
|
72 |
gr.Examples(
|
73 |
[
|
@@ -87,13 +92,13 @@ with gr.Blocks(css="app.css") as demo:
|
|
87 |
submit.click(
|
88 |
fn=preprocess_image_and_predict,
|
89 |
inputs=[input_image],
|
90 |
-
outputs=[output_image, output_label],
|
91 |
queue=True,
|
92 |
)
|
93 |
clear_btn.click(
|
94 |
fn=clear_static_info,
|
95 |
inputs=[],
|
96 |
-
outputs=[input_image, output_image, output_label],
|
97 |
queue=True,
|
98 |
)
|
99 |
|
@@ -102,7 +107,8 @@ with gr.Blocks(css="app.css") as demo:
|
|
102 |
inputs=input_video,
|
103 |
outputs=[
|
104 |
output_video,
|
105 |
-
output_face,
|
|
|
106 |
output_statistics
|
107 |
],
|
108 |
queue=True,
|
@@ -113,7 +119,8 @@ with gr.Blocks(css="app.css") as demo:
|
|
113 |
outputs=[
|
114 |
input_video,
|
115 |
output_video,
|
116 |
-
output_face,
|
|
|
117 |
output_statistics
|
118 |
],
|
119 |
queue=True,
|
|
|
18 |
def clear_static_info():
|
19 |
return (
|
20 |
gr.Image(value=None, type="pil"),
|
21 |
+
gr.Image(value=None, scale=1, elem_classes="dl5"),
|
22 |
gr.Image(value=None, scale=1, elem_classes="dl2"),
|
23 |
gr.Label(value=None, num_top_classes=3, scale=1, elem_classes="dl3"),
|
24 |
)
|
|
|
28 |
gr.Video(value=None),
|
29 |
gr.Video(value=None),
|
30 |
gr.Video(value=None),
|
31 |
+
gr.Video(value=None),
|
32 |
gr.Plot(value=None),
|
33 |
)
|
34 |
|
|
|
47 |
)
|
48 |
with gr.Column(scale=2, elem_classes="dl4"):
|
49 |
with gr.Row():
|
50 |
+
output_video = gr.Video(label="Original video", scale=1, elem_classes="video2")
|
51 |
output_face = gr.Video(label="Pre-processed video", scale=1, elem_classes="video3")
|
52 |
+
output_heatmaps = gr.Video(label="Heatmaps", scale=1, elem_classes="video4")
|
53 |
output_statistics = gr.Plot(label="Statistics of emotions", elem_classes="stat")
|
54 |
gr.Examples(
|
55 |
["videos/video1.mp4",
|
|
|
61 |
gr.Markdown(value=DESCRIPTION_STATIC)
|
62 |
with gr.Row():
|
63 |
with gr.Column(scale=2, elem_classes="dl1"):
|
64 |
+
input_image = gr.Image(label="Original image", type="pil")
|
65 |
with gr.Row():
|
66 |
clear_btn = gr.Button(
|
67 |
value="Clear", interactive=True, scale=1, elem_classes="clear"
|
|
|
70 |
value="Submit", interactive=True, scale=1, elem_classes="submit"
|
71 |
)
|
72 |
with gr.Column(scale=1, elem_classes="dl4"):
|
73 |
+
with gr.Row():
|
74 |
+
output_image = gr.Image(label="Face", scale=1, elem_classes="dl5")
|
75 |
+
output_heatmap = gr.Image(label="Heatmap", scale=1, elem_classes="dl2")
|
76 |
output_label = gr.Label(num_top_classes=3, scale=1, elem_classes="dl3")
|
77 |
gr.Examples(
|
78 |
[
|
|
|
92 |
submit.click(
|
93 |
fn=preprocess_image_and_predict,
|
94 |
inputs=[input_image],
|
95 |
+
outputs=[output_image, output_heatmap, output_label],
|
96 |
queue=True,
|
97 |
)
|
98 |
clear_btn.click(
|
99 |
fn=clear_static_info,
|
100 |
inputs=[],
|
101 |
+
outputs=[input_image, output_image, output_heatmap, output_label],
|
102 |
queue=True,
|
103 |
)
|
104 |
|
|
|
107 |
inputs=input_video,
|
108 |
outputs=[
|
109 |
output_video,
|
110 |
+
output_face,
|
111 |
+
output_heatmaps,
|
112 |
output_statistics
|
113 |
],
|
114 |
queue=True,
|
|
|
119 |
outputs=[
|
120 |
input_video,
|
121 |
output_video,
|
122 |
+
output_face,
|
123 |
+
output_heatmaps,
|
124 |
output_statistics
|
125 |
],
|
126 |
queue=True,
|
app/app_utils.py
CHANGED
@@ -10,9 +10,10 @@ import numpy as np
|
|
10 |
import mediapipe as mp
|
11 |
from PIL import Image
|
12 |
import cv2
|
|
|
13 |
|
14 |
# Importing necessary components for the Gradio app
|
15 |
-
from app.model import pth_model_static, pth_model_dynamic, pth_processing
|
16 |
from app.face_utils import get_box, display_info
|
17 |
from app.config import DICT_EMO, config_data
|
18 |
from app.plot import statistics_plot
|
@@ -49,8 +50,13 @@ def preprocess_image_and_predict(inp):
|
|
49 |
.numpy()[0]
|
50 |
)
|
51 |
confidences = {DICT_EMO[i]: float(prediction[i]) for i in range(7)}
|
|
|
|
|
|
|
|
|
|
|
52 |
|
53 |
-
return cur_face, confidences
|
54 |
|
55 |
|
56 |
def preprocess_video_and_predict(video):
|
@@ -60,14 +66,20 @@ def preprocess_video_and_predict(video):
|
|
60 |
h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
61 |
fps = np.round(cap.get(cv2.CAP_PROP_FPS))
|
62 |
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
65 |
|
66 |
lstm_features = []
|
67 |
count_frame = 1
|
|
|
68 |
probs = []
|
69 |
frames = []
|
70 |
-
last_output = None
|
|
|
|
|
71 |
|
72 |
with mp_face_mesh.FaceMesh(
|
73 |
max_num_faces=1,
|
@@ -90,9 +102,16 @@ def preprocess_video_and_predict(video):
|
|
90 |
startX, startY, endX, endY = get_box(fl, w, h)
|
91 |
cur_face = frame_copy[startY:endY, startX: endX]
|
92 |
|
93 |
-
if
|
94 |
cur_face_copy = pth_processing(Image.fromarray(cur_face))
|
95 |
features = torch.nn.functional.relu(pth_model_static.extract_features(cur_face_copy)).detach().numpy()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
|
97 |
if len(lstm_features) == 0:
|
98 |
lstm_features = [features]*10
|
@@ -103,32 +122,48 @@ def preprocess_video_and_predict(video):
|
|
103 |
lstm_f = torch.unsqueeze(lstm_f, 0)
|
104 |
output = pth_model_dynamic(lstm_f).detach().numpy()
|
105 |
last_output = output
|
|
|
|
|
|
|
|
|
106 |
else:
|
107 |
if last_output is not None:
|
108 |
output = last_output
|
|
|
|
|
109 |
elif last_output is None:
|
110 |
-
output = np.
|
|
|
111 |
|
112 |
probs.append(output[0])
|
113 |
frames.append(count_frame)
|
114 |
else:
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
|
123 |
-
cur_face = display_info(cur_face, 'Frame: {}'.format(count_frame), box_scale=.3)
|
124 |
count_frame += 1
|
125 |
-
|
|
|
126 |
|
127 |
-
|
|
|
128 |
|
129 |
stat = statistics_plot(frames, probs)
|
130 |
|
131 |
if not stat:
|
132 |
-
return None, None, None
|
133 |
|
134 |
-
return video,
|
|
|
10 |
import mediapipe as mp
|
11 |
from PIL import Image
|
12 |
import cv2
|
13 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image
|
14 |
|
15 |
# Importing necessary components for the Gradio app
|
16 |
+
from app.model import pth_model_static, pth_model_dynamic, cam, pth_processing
|
17 |
from app.face_utils import get_box, display_info
|
18 |
from app.config import DICT_EMO, config_data
|
19 |
from app.plot import statistics_plot
|
|
|
50 |
.numpy()[0]
|
51 |
)
|
52 |
confidences = {DICT_EMO[i]: float(prediction[i]) for i in range(7)}
|
53 |
+
grayscale_cam = cam(input_tensor=cur_face_n)
|
54 |
+
grayscale_cam = grayscale_cam[0, :]
|
55 |
+
cur_face_hm = cv2.resize(cur_face,(224,224))
|
56 |
+
cur_face_hm = np.float32(cur_face_hm) / 255
|
57 |
+
heatmap = show_cam_on_image(cur_face_hm, grayscale_cam, use_rgb=True)
|
58 |
|
59 |
+
return cur_face, heatmap, confidences
|
60 |
|
61 |
|
62 |
def preprocess_video_and_predict(video):
|
|
|
66 |
h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
67 |
fps = np.round(cap.get(cv2.CAP_PROP_FPS))
|
68 |
|
69 |
+
path_save_video_face = 'result_face.mp4'
|
70 |
+
vid_writer_face = cv2.VideoWriter(path_save_video_face, cv2.VideoWriter_fourcc(*'mp4v'), fps, (224, 224))
|
71 |
+
|
72 |
+
path_save_video_hm = 'result_hm.mp4'
|
73 |
+
vid_writer_hm = cv2.VideoWriter(path_save_video_hm, cv2.VideoWriter_fourcc(*'mp4v'), fps, (224, 224))
|
74 |
|
75 |
lstm_features = []
|
76 |
count_frame = 1
|
77 |
+
count_face = 0
|
78 |
probs = []
|
79 |
frames = []
|
80 |
+
last_output = None
|
81 |
+
last_heatmap = None
|
82 |
+
cur_face = None
|
83 |
|
84 |
with mp_face_mesh.FaceMesh(
|
85 |
max_num_faces=1,
|
|
|
102 |
startX, startY, endX, endY = get_box(fl, w, h)
|
103 |
cur_face = frame_copy[startY:endY, startX: endX]
|
104 |
|
105 |
+
if count_face%config_data.FRAME_DOWNSAMPLING == 0:
|
106 |
cur_face_copy = pth_processing(Image.fromarray(cur_face))
|
107 |
features = torch.nn.functional.relu(pth_model_static.extract_features(cur_face_copy)).detach().numpy()
|
108 |
+
|
109 |
+
grayscale_cam = cam(input_tensor=cur_face_copy)
|
110 |
+
grayscale_cam = grayscale_cam[0, :]
|
111 |
+
cur_face_hm = cv2.resize(cur_face,(224,224), interpolation = cv2.INTER_AREA)
|
112 |
+
cur_face_hm = np.float32(cur_face_hm) / 255
|
113 |
+
heatmap = show_cam_on_image(cur_face_hm, grayscale_cam, use_rgb=False)
|
114 |
+
last_heatmap = heatmap
|
115 |
|
116 |
if len(lstm_features) == 0:
|
117 |
lstm_features = [features]*10
|
|
|
122 |
lstm_f = torch.unsqueeze(lstm_f, 0)
|
123 |
output = pth_model_dynamic(lstm_f).detach().numpy()
|
124 |
last_output = output
|
125 |
+
|
126 |
+
if count_face == 0:
|
127 |
+
count_face += 1
|
128 |
+
|
129 |
else:
|
130 |
if last_output is not None:
|
131 |
output = last_output
|
132 |
+
heatmap = last_heatmap
|
133 |
+
|
134 |
elif last_output is None:
|
135 |
+
output = np.empty((1, 7))
|
136 |
+
output[:] = np.nan
|
137 |
|
138 |
probs.append(output[0])
|
139 |
frames.append(count_frame)
|
140 |
else:
|
141 |
+
if last_output is not None:
|
142 |
+
lstm_features = []
|
143 |
+
empty = np.empty((7))
|
144 |
+
empty[:] = np.nan
|
145 |
+
probs.append(empty)
|
146 |
+
frames.append(count_frame)
|
147 |
+
|
148 |
+
if cur_face is not None:
|
149 |
+
heatmap_f = display_info(heatmap, 'Frame: {}'.format(count_frame), box_scale=.3)
|
150 |
+
|
151 |
+
cur_face = cv2.cvtColor(cur_face, cv2.COLOR_RGB2BGR)
|
152 |
+
cur_face = cv2.resize(cur_face, (224,224), interpolation = cv2.INTER_AREA)
|
153 |
+
cur_face = display_info(cur_face, 'Frame: {}'.format(count_frame), box_scale=.3)
|
154 |
+
vid_writer_face.write(cur_face)
|
155 |
+
vid_writer_hm.write(heatmap_f)
|
156 |
|
|
|
157 |
count_frame += 1
|
158 |
+
if count_face != 0:
|
159 |
+
count_face += 1
|
160 |
|
161 |
+
vid_writer_face.release()
|
162 |
+
vid_writer_hm.release()
|
163 |
|
164 |
stat = statistics_plot(frames, probs)
|
165 |
|
166 |
if not stat:
|
167 |
+
return None, None, None, None
|
168 |
|
169 |
+
return video, path_save_video_face, path_save_video_hm, stat
|
app/face_utils.py
CHANGED
@@ -34,7 +34,8 @@ def get_box(fl, w, h):
|
|
34 |
return startX, startY, endX, endY
|
35 |
|
36 |
def display_info(img, text, margin=1.0, box_scale=1.0):
|
37 |
-
|
|
|
38 |
line_width = int(min(img_h, img_w) * 0.001)
|
39 |
thickness = max(int(line_width / 3), 1)
|
40 |
|
@@ -45,15 +46,15 @@ def display_info(img, text, margin=1.0, box_scale=1.0):
|
|
45 |
t_w, t_h = cv2.getTextSize(text, font_face, font_scale, None)[0]
|
46 |
|
47 |
margin_n = int(t_h * margin)
|
48 |
-
sub_img =
|
49 |
img_w - t_w - margin_n - int(2 * t_h * box_scale): img_w - margin_n]
|
50 |
|
51 |
white_rect = np.ones(sub_img.shape, dtype=np.uint8) * 255
|
52 |
|
53 |
-
|
54 |
img_w - t_w - margin_n - int(2 * t_h * box_scale):img_w - margin_n] = cv2.addWeighted(sub_img, 0.5, white_rect, .5, 1.0)
|
55 |
|
56 |
-
cv2.putText(img=
|
57 |
text=text,
|
58 |
org=(img_w - t_w - margin_n - int(2 * t_h * box_scale) // 2,
|
59 |
0 + margin_n + t_h + int(2 * t_h * box_scale) // 2),
|
@@ -64,4 +65,4 @@ def display_info(img, text, margin=1.0, box_scale=1.0):
|
|
64 |
lineType=cv2.LINE_AA,
|
65 |
bottomLeftOrigin=False)
|
66 |
|
67 |
-
return
|
|
|
34 |
return startX, startY, endX, endY
|
35 |
|
36 |
def display_info(img, text, margin=1.0, box_scale=1.0):
|
37 |
+
img_copy = img.copy()
|
38 |
+
img_h, img_w, _ = img_copy.shape
|
39 |
line_width = int(min(img_h, img_w) * 0.001)
|
40 |
thickness = max(int(line_width / 3), 1)
|
41 |
|
|
|
46 |
t_w, t_h = cv2.getTextSize(text, font_face, font_scale, None)[0]
|
47 |
|
48 |
margin_n = int(t_h * margin)
|
49 |
+
sub_img = img_copy[0 + margin_n: 0 + margin_n + t_h + int(2 * t_h * box_scale),
|
50 |
img_w - t_w - margin_n - int(2 * t_h * box_scale): img_w - margin_n]
|
51 |
|
52 |
white_rect = np.ones(sub_img.shape, dtype=np.uint8) * 255
|
53 |
|
54 |
+
img_copy[0 + margin_n: 0 + margin_n + t_h + int(2 * t_h * box_scale),
|
55 |
img_w - t_w - margin_n - int(2 * t_h * box_scale):img_w - margin_n] = cv2.addWeighted(sub_img, 0.5, white_rect, .5, 1.0)
|
56 |
|
57 |
+
cv2.putText(img=img_copy,
|
58 |
text=text,
|
59 |
org=(img_w - t_w - margin_n - int(2 * t_h * box_scale) // 2,
|
60 |
0 + margin_n + t_h + int(2 * t_h * box_scale) // 2),
|
|
|
65 |
lineType=cv2.LINE_AA,
|
66 |
bottomLeftOrigin=False)
|
67 |
|
68 |
+
return img_copy
|
app/model.py
CHANGED
@@ -10,9 +10,11 @@ import torch
|
|
10 |
import requests
|
11 |
from PIL import Image
|
12 |
from torchvision import transforms
|
|
|
13 |
|
14 |
# Importing necessary components for the Gradio app
|
15 |
from app.config import config_data
|
|
|
16 |
|
17 |
|
18 |
def load_model(model_url, model_path):
|
@@ -21,17 +23,23 @@ def load_model(model_url, model_path):
|
|
21 |
with open(model_path, "wb") as file:
|
22 |
for chunk in response.iter_content(chunk_size=8192):
|
23 |
file.write(chunk)
|
24 |
-
return
|
25 |
except Exception as e:
|
26 |
print(f"Error loading model: {e}")
|
27 |
return None
|
28 |
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
pth_model_dynamic
|
33 |
-
|
34 |
|
|
|
|
|
35 |
|
36 |
def pth_processing(fp):
|
37 |
class PreprocessInput(torch.nn.Module):
|
|
|
10 |
import requests
|
11 |
from PIL import Image
|
12 |
from torchvision import transforms
|
13 |
+
from pytorch_grad_cam import GradCAM
|
14 |
|
15 |
# Importing necessary components for the Gradio app
|
16 |
from app.config import config_data
|
17 |
+
from app.model_architectures import ResNet50, LSTMPyTorch
|
18 |
|
19 |
|
20 |
def load_model(model_url, model_path):
|
|
|
23 |
with open(model_path, "wb") as file:
|
24 |
for chunk in response.iter_content(chunk_size=8192):
|
25 |
file.write(chunk)
|
26 |
+
return model_path
|
27 |
except Exception as e:
|
28 |
print(f"Error loading model: {e}")
|
29 |
return None
|
30 |
|
31 |
+
path_static = load_model(config_data.model_static_url, config_data.model_static_path)
|
32 |
+
pth_model_static = ResNet50(7, channels=3)
|
33 |
+
pth_model_static.load_state_dict(torch.load(path_static))
|
34 |
+
pth_model_static.eval()
|
35 |
|
36 |
+
path_dynamic = load_model(config_data.model_dynamic_url, config_data.model_dynamic_path)
|
37 |
+
pth_model_dynamic = LSTMPyTorch()
|
38 |
+
pth_model_dynamic.load_state_dict(torch.load(path_dynamic))
|
39 |
+
pth_model_dynamic.eval()
|
40 |
|
41 |
+
target_layers = [pth_model_static.layer4]
|
42 |
+
cam = GradCAM(model=pth_model_static, target_layers=target_layers)
|
43 |
|
44 |
def pth_processing(fp):
|
45 |
class PreprocessInput(torch.nn.Module):
|
app/model_architectures.py
ADDED
@@ -0,0 +1,150 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
File: model.py
|
3 |
+
Author: Elena Ryumina and Dmitry Ryumin
|
4 |
+
Description: This module provides model architectures.
|
5 |
+
License: MIT License
|
6 |
+
"""
|
7 |
+
|
8 |
+
import torch
|
9 |
+
import torch.nn as nn
|
10 |
+
import torch.nn.functional as F
|
11 |
+
import math
|
12 |
+
|
13 |
+
class Bottleneck(nn.Module):
|
14 |
+
expansion = 4
|
15 |
+
def __init__(self, in_channels, out_channels, i_downsample=None, stride=1):
|
16 |
+
super(Bottleneck, self).__init__()
|
17 |
+
|
18 |
+
self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=stride, padding=0, bias=False)
|
19 |
+
self.batch_norm1 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.99)
|
20 |
+
|
21 |
+
self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding='same', bias=False)
|
22 |
+
self.batch_norm2 = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.99)
|
23 |
+
|
24 |
+
self.conv3 = nn.Conv2d(out_channels, out_channels*self.expansion, kernel_size=1, stride=1, padding=0, bias=False)
|
25 |
+
self.batch_norm3 = nn.BatchNorm2d(out_channels*self.expansion, eps=0.001, momentum=0.99)
|
26 |
+
|
27 |
+
self.i_downsample = i_downsample
|
28 |
+
self.stride = stride
|
29 |
+
self.relu = nn.ReLU()
|
30 |
+
|
31 |
+
def forward(self, x):
|
32 |
+
identity = x.clone()
|
33 |
+
x = self.relu(self.batch_norm1(self.conv1(x)))
|
34 |
+
|
35 |
+
x = self.relu(self.batch_norm2(self.conv2(x)))
|
36 |
+
|
37 |
+
x = self.conv3(x)
|
38 |
+
x = self.batch_norm3(x)
|
39 |
+
|
40 |
+
#downsample if needed
|
41 |
+
if self.i_downsample is not None:
|
42 |
+
identity = self.i_downsample(identity)
|
43 |
+
#add identity
|
44 |
+
x+=identity
|
45 |
+
x=self.relu(x)
|
46 |
+
|
47 |
+
return x
|
48 |
+
|
49 |
+
class Conv2dSame(torch.nn.Conv2d):
|
50 |
+
|
51 |
+
def calc_same_pad(self, i: int, k: int, s: int, d: int) -> int:
|
52 |
+
return max((math.ceil(i / s) - 1) * s + (k - 1) * d + 1 - i, 0)
|
53 |
+
|
54 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
55 |
+
ih, iw = x.size()[-2:]
|
56 |
+
|
57 |
+
pad_h = self.calc_same_pad(i=ih, k=self.kernel_size[0], s=self.stride[0], d=self.dilation[0])
|
58 |
+
pad_w = self.calc_same_pad(i=iw, k=self.kernel_size[1], s=self.stride[1], d=self.dilation[1])
|
59 |
+
|
60 |
+
if pad_h > 0 or pad_w > 0:
|
61 |
+
x = F.pad(
|
62 |
+
x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2]
|
63 |
+
)
|
64 |
+
return F.conv2d(
|
65 |
+
x,
|
66 |
+
self.weight,
|
67 |
+
self.bias,
|
68 |
+
self.stride,
|
69 |
+
self.padding,
|
70 |
+
self.dilation,
|
71 |
+
self.groups,
|
72 |
+
)
|
73 |
+
|
74 |
+
class ResNet(nn.Module):
|
75 |
+
def __init__(self, ResBlock, layer_list, num_classes, num_channels=3):
|
76 |
+
super(ResNet, self).__init__()
|
77 |
+
self.in_channels = 64
|
78 |
+
|
79 |
+
self.conv_layer_s2_same = Conv2dSame(num_channels, 64, 7, stride=2, groups=1, bias=False)
|
80 |
+
self.batch_norm1 = nn.BatchNorm2d(64, eps=0.001, momentum=0.99)
|
81 |
+
self.relu = nn.ReLU()
|
82 |
+
self.max_pool = nn.MaxPool2d(kernel_size = 3, stride=2)
|
83 |
+
|
84 |
+
self.layer1 = self._make_layer(ResBlock, layer_list[0], planes=64, stride=1)
|
85 |
+
self.layer2 = self._make_layer(ResBlock, layer_list[1], planes=128, stride=2)
|
86 |
+
self.layer3 = self._make_layer(ResBlock, layer_list[2], planes=256, stride=2)
|
87 |
+
self.layer4 = self._make_layer(ResBlock, layer_list[3], planes=512, stride=2)
|
88 |
+
|
89 |
+
self.avgpool = nn.AdaptiveAvgPool2d((1,1))
|
90 |
+
self.fc1 = nn.Linear(512*ResBlock.expansion, 512)
|
91 |
+
self.relu1 = nn.ReLU()
|
92 |
+
self.fc2 = nn.Linear(512, num_classes)
|
93 |
+
|
94 |
+
def extract_features(self, x):
|
95 |
+
x = self.relu(self.batch_norm1(self.conv_layer_s2_same(x)))
|
96 |
+
x = self.max_pool(x)
|
97 |
+
# print(x.shape)
|
98 |
+
x = self.layer1(x)
|
99 |
+
x = self.layer2(x)
|
100 |
+
x = self.layer3(x)
|
101 |
+
x = self.layer4(x)
|
102 |
+
|
103 |
+
x = self.avgpool(x)
|
104 |
+
x = x.reshape(x.shape[0], -1)
|
105 |
+
x = self.fc1(x)
|
106 |
+
return x
|
107 |
+
|
108 |
+
def forward(self, x):
|
109 |
+
x = self.extract_features(x)
|
110 |
+
x = self.relu1(x)
|
111 |
+
x = self.fc2(x)
|
112 |
+
return x
|
113 |
+
|
114 |
+
def _make_layer(self, ResBlock, blocks, planes, stride=1):
|
115 |
+
ii_downsample = None
|
116 |
+
layers = []
|
117 |
+
|
118 |
+
if stride != 1 or self.in_channels != planes*ResBlock.expansion:
|
119 |
+
ii_downsample = nn.Sequential(
|
120 |
+
nn.Conv2d(self.in_channels, planes*ResBlock.expansion, kernel_size=1, stride=stride, bias=False, padding=0),
|
121 |
+
nn.BatchNorm2d(planes*ResBlock.expansion, eps=0.001, momentum=0.99)
|
122 |
+
)
|
123 |
+
|
124 |
+
layers.append(ResBlock(self.in_channels, planes, i_downsample=ii_downsample, stride=stride))
|
125 |
+
self.in_channels = planes*ResBlock.expansion
|
126 |
+
|
127 |
+
for i in range(blocks-1):
|
128 |
+
layers.append(ResBlock(self.in_channels, planes))
|
129 |
+
|
130 |
+
return nn.Sequential(*layers)
|
131 |
+
|
132 |
+
def ResNet50(num_classes, channels=3):
|
133 |
+
return ResNet(Bottleneck, [3,4,6,3], num_classes, channels)
|
134 |
+
|
135 |
+
|
136 |
+
class LSTMPyTorch(nn.Module):
|
137 |
+
def __init__(self):
|
138 |
+
super(LSTMPyTorch, self).__init__()
|
139 |
+
|
140 |
+
self.lstm1 = nn.LSTM(input_size=512, hidden_size=512, batch_first=True, bidirectional=False)
|
141 |
+
self.lstm2 = nn.LSTM(input_size=512, hidden_size=256, batch_first=True, bidirectional=False)
|
142 |
+
self.fc = nn.Linear(256, 7)
|
143 |
+
self.softmax = nn.Softmax(dim=1)
|
144 |
+
|
145 |
+
def forward(self, x):
|
146 |
+
x, _ = self.lstm1(x)
|
147 |
+
x, _ = self.lstm2(x)
|
148 |
+
x = self.fc(x[:, -1, :])
|
149 |
+
x = self.softmax(x)
|
150 |
+
return x
|
config.toml
CHANGED
@@ -2,9 +2,9 @@ APP_VERSION = "0.2.0"
|
|
2 |
FRAME_DOWNSAMPLING = 5
|
3 |
|
4 |
[model_static]
|
5 |
-
url = "https://huggingface.co/ElenaRyumina/face_emotion_recognition/resolve/main/FER_static_ResNet50_AffectNet.
|
6 |
-
path = "FER_static_ResNet50_AffectNet.
|
7 |
|
8 |
[model_dynamic]
|
9 |
-
url = "https://huggingface.co/ElenaRyumina/face_emotion_recognition/resolve/main/FER_dinamic_LSTM_IEMOCAP.
|
10 |
-
path = "FER_dinamic_LSTM_IEMOCAP.
|
|
|
2 |
FRAME_DOWNSAMPLING = 5
|
3 |
|
4 |
[model_static]
|
5 |
+
url = "https://huggingface.co/ElenaRyumina/face_emotion_recognition/resolve/main/FER_static_ResNet50_AffectNet.pt"
|
6 |
+
path = "FER_static_ResNet50_AffectNet.pt"
|
7 |
|
8 |
[model_dynamic]
|
9 |
+
url = "https://huggingface.co/ElenaRyumina/face_emotion_recognition/resolve/main/FER_dinamic_LSTM_IEMOCAP.pt"
|
10 |
+
path = "FER_dinamic_LSTM_IEMOCAP.pt"
|
result.mp4
DELETED
Binary file (108 kB)
|
|