Timilehin commited on
Commit
5271a4b
·
verified ·
1 Parent(s): 8756727

Upload main1.py

Browse files
Files changed (1) hide show
  1. main1.py +288 -0
main1.py ADDED
@@ -0,0 +1,288 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from anchor_utils import *
2
+ import os
3
+ import torch
4
+ import cv2
5
+ import numpy as np
6
+ from PIL import Image
7
+ from flask import Flask, request, send_file, make_response
8
+ from io import BytesIO
9
+
10
+ from flask import Flask, request, jsonify, render_template
11
+ from rembg import remove
12
+ import subprocess
13
+ import shutil
14
+
15
+ # Set up paths
16
+
17
+ # image_dir = os.path.join(os.getcwd(), 'pifuhd', 'sample_images'
18
+ # output_dir = os.path.join(os.getcwd(), "pifuhd", "results")
19
+ ckpt_path = os.path.join(os.getcwd(), 'pifuhd', 'checkpoints', 'pifuhd.pt')
20
+
21
+ import io
22
+
23
+
24
+ # os.chdir(os.path.join(os.getcwd(), 'lightweight-human-pose-estimation.pytorch'))
25
+ print('cwd', os.getcwd())
26
+ print(os.listdir(os.getcwd()))
27
+
28
+ import sys
29
+ sys.path.append(os.path.join(os.getcwd(), 'lightweight-human-pose-estimation.pytorch'))
30
+ # Import necessary modules from lightweight human pose estimation
31
+ from models.with_mobilenet import PoseEstimationWithMobileNet
32
+ from modules.keypoints import extract_keypoints, group_keypoints
33
+ from modules.load_state import load_state
34
+ from modules.pose import Pose
35
+ import demo
36
+
37
+ # Initialize Flask app
38
+ app = Flask(__name__)
39
+
40
+ UPLOAD_FOLDER = os.getenv('UPLOAD_FOLDER', '/tmp/sample_images')
41
+ app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER
42
+
43
+ if not os.path.exists(app.config['UPLOAD_FOLDER']):
44
+ os.makedirs(app.config['UPLOAD_FOLDER'])
45
+
46
+ image_dir = os.path.join(app.config['UPLOAD_FOLDER'])
47
+ print('img_dir', image_dir)
48
+
49
+ # Load pose estimation model
50
+ net = PoseEstimationWithMobileNet()
51
+ checkpoint = torch.load(os.path.join(os.getcwd(), "lightweight-human-pose-estimation.pytorch", "checkpoint_iter_370000.pth"), map_location='cpu')
52
+ load_state(net, checkpoint)
53
+ net = net.eval()
54
+
55
+
56
+ # Preprocess function to crop image
57
+ def get_rect(net, image_path, height_size=512):
58
+ stride = 8
59
+ upsample_ratio = 4
60
+ num_keypoints = Pose.num_kpts
61
+ img = cv2.imread(image_path, cv2.IMREAD_COLOR)
62
+ orig_img = img.copy()
63
+ heatmaps, pafs, scale, pad = demo.infer_fast(net, img, height_size, stride, upsample_ratio, cpu=True)
64
+
65
+ total_keypoints_num = 0
66
+ all_keypoints_by_type = []
67
+ for kpt_idx in range(num_keypoints):
68
+ total_keypoints_num += extract_keypoints(heatmaps[:, :, kpt_idx], all_keypoints_by_type, total_keypoints_num)
69
+
70
+ pose_entries, all_keypoints = group_keypoints(all_keypoints_by_type, pafs)
71
+ for kpt_id in range(all_keypoints.shape[0]):
72
+ all_keypoints[kpt_id, 0] = (all_keypoints[kpt_id, 0] * stride / upsample_ratio - pad[1]) / scale
73
+ all_keypoints[kpt_id, 1] = (all_keypoints[kpt_id, 1] * stride / upsample_ratio - pad[0]) / scale
74
+
75
+ rects = []
76
+ for n in range(len(pose_entries)):
77
+ if len(pose_entries[n]) == 0:
78
+ continue
79
+ pose_keypoints = np.ones((num_keypoints, 2), dtype=np.int32) * -1
80
+ valid_keypoints = []
81
+ for kpt_id in range(num_keypoints):
82
+ if pose_entries[n][kpt_id] != -1.0:
83
+ pose_keypoints[kpt_id, 0] = int(all_keypoints[int(pose_entries[n][kpt_id]), 0])
84
+ pose_keypoints[kpt_id, 1] = int(all_keypoints[int(pose_entries[n][kpt_id]), 1])
85
+ valid_keypoints.append([pose_keypoints[kpt_id, 0], pose_keypoints[kpt_id, 1]])
86
+ valid_keypoints = np.array(valid_keypoints)
87
+
88
+ if pose_entries[n][10] != -1.0 or pose_entries[n][13] != -1.0:
89
+ pmin = valid_keypoints.min(0)
90
+ pmax = valid_keypoints.max(0)
91
+ center = (0.5 * (pmax[:2] + pmin[:2])).astype(np.int32)
92
+ radius = int(0.65 * max(pmax[0] - pmin[0], pmax[1] - pmin[1]))
93
+ elif pose_entries[n][10] == -1.0 and pose_entries[n][13] == -1.0 and pose_entries[n][8] != -1.0 and pose_entries[n][11] != -1.0:
94
+ center = (0.5 * (pose_keypoints[8] + pose_keypoints[11])).astype(np.int32)
95
+ radius = int(1.45 * np.sqrt(((center[None, :] - valid_keypoints) ** 2).sum(1)).max(0))
96
+ center[1] += int(0.05 * radius)
97
+ else:
98
+ center = np.array([img.shape[1] // 2, img.shape[0] // 2])
99
+ radius = max(img.shape[1] // 2, img.shape[0] // 2)
100
+
101
+ x1 = center[0] - radius
102
+ y1 = center[1] - radius
103
+
104
+ rects.append([x1, y1, 2 * radius, 2 * radius])
105
+
106
+ return np.array(rects)
107
+
108
+
109
+
110
+ from pifuhd.apps.simple_test import run_pifuhd
111
+
112
+ def process_image_pifuhd(image):
113
+ # Define paths
114
+ # image_path = os.path.join(os.getcwd(), image_dir, "uploaded_image.png")
115
+ image_path = os.path.join(app.config['UPLOAD_FOLDER'], "uploaded_image.png")
116
+
117
+ # obj_path = os.path.join(output_dir, "output.obj")
118
+
119
+ # Save the uploaded image
120
+ image.save(image_path)
121
+
122
+ # Preprocess the image to get rect
123
+ rects = get_rect(net, image_path)
124
+ rect_path = image_path.replace('.png', '_rect.txt')
125
+ np.savetxt(rect_path, rects, fmt='%d')
126
+
127
+ # Run PIFuHD and get the buffer
128
+ obj_buffer = run_pifuhd(
129
+ input_path=image_dir,
130
+ ckpt_path=ckpt_path,
131
+ resolution=256,
132
+ use_rect=True
133
+ )
134
+
135
+ # # # Save the first buffer as an OBJ file
136
+ # with open(obj_path, 'wb') as f:
137
+ # f.write(obj_buffer[0].getvalue())
138
+
139
+ return obj_buffer[0]
140
+
141
+
142
+ @app.route('/')
143
+ def pifuhd_page():
144
+ return render_template('pifuhd.html')
145
+
146
+ @app.route('/generate-mesh', methods=['POST'])
147
+ def generate_mesh():
148
+ if 'file' not in request.files:
149
+ return "No file part", 400
150
+
151
+ file = request.files['file']
152
+ if file.filename == '':
153
+ return "No selected file", 400
154
+
155
+ if file:
156
+ # Convert the file to an image and process it
157
+ image = Image.open(file.stream)
158
+ obj_buffer = process_image_pifuhd(image)
159
+
160
+ # Create a BytesIO object to send the file
161
+ obj_bytes_io = io.BytesIO(obj_buffer.getvalue())
162
+ obj_bytes_io.seek(0) # Ensure the buffer is at the start
163
+
164
+ # Return the OBJ file in memory
165
+ return send_file(
166
+ obj_bytes_io,
167
+ as_attachment=True,
168
+ download_name='output.obj',
169
+ mimetype='application/octet-stream'
170
+ )
171
+
172
+
173
+
174
+
175
+
176
+
177
+ # Helper function to process images
178
+ def process_image(filename):
179
+ # input_path = f'{os.getcwd()}/sample_images/{filename}'
180
+ input_path = filename
181
+ output_path = f'{os.getcwd()}/AnchorUDF/pics/my_test/RENDER/1/0_0_00.jpg'
182
+
183
+ input_image = Image.open(input_path)
184
+
185
+ # Determine new size
186
+ if input_image.size[0] <= 600 or input_image.size[1] <= 600:
187
+ new_size = (500, 500)
188
+ elif input_image.size[0] >= 900 or input_image.size[1] >= 900:
189
+ new_size = (1500, 1500)
190
+ elif input_image.size[0] <= 900 or input_image.size[1] <= 900 and input_image.size[0] >= 700 or input_image.size[1] >= 700:
191
+ new_size = (1000, 1000)
192
+ else:
193
+ new_size = input_image.size
194
+
195
+ resized_image = input_image.resize(new_size, Image.LANCZOS)
196
+
197
+ output_image = remove(resized_image)
198
+ rgb_output_image = output_image.convert('RGB')
199
+
200
+ # Save the processed image
201
+ os.makedirs(os.path.dirname(output_path), exist_ok=True)
202
+ rgb_output_image.save(output_path, quality=200, optimize=True)
203
+
204
+ return output_path
205
+
206
+ # Helper function to generate mask
207
+ def generate_mask(image_path):
208
+ image = cv2.imread(image_path)
209
+ gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
210
+ _, binary = cv2.threshold(gray, 1, 255, cv2.THRESH_BINARY)
211
+ contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
212
+ mask = np.zeros_like(gray)
213
+ cv2.drawContours(mask, contours, -1, 255, thickness=cv2.FILLED)
214
+
215
+ mask_path = f"{os.getcwd()}/AnchorUDF/pics/my_test/MASK/1/0_0_00.png"
216
+ cv2.imwrite(mask_path, mask)
217
+
218
+ return mask_path
219
+
220
+
221
+ # Helper function for mesh processing and returning cleaned .ply file
222
+ def process_mesh(scale_factor=1.0, nb_neighbors=40, std_ratio=1.0, voxel_size=0.005):
223
+ off_path = None
224
+ al_obj = glob.glob(f'{os.getcwd()}/AnchorUDF/out/out/*.off')
225
+ # print('all_obg', al_obj)
226
+ for i in al_obj:
227
+ if os.path.getsize(i) > 30000:
228
+ print('found off file', i)
229
+ off_path = i
230
+ break
231
+
232
+ if not off_path:
233
+ raise FileNotFoundError("No suitable OFF file found.")
234
+
235
+ filename = 'output'
236
+ saved_ply_path = f"{off_path}_{filename}.ply"
237
+
238
+ convert_off_to_ply(off_path, saved_ply_path, scale_factor)
239
+
240
+ cleaned_ply_path, cleaned_pcd_path = remove_outliers_and_convert_to_point_cloud(saved_ply_path, nb_neighbors, std_ratio, voxel_size)
241
+
242
+ return cleaned_ply_path, cleaned_pcd_path
243
+
244
+ # Helper function for garment inference
245
+ def run_garment_inference():
246
+ # os.makedirs("out", exist_ok=True)
247
+ print('garm working dir', os.getcwd())
248
+ shutil.rmtree(os.path.join(os.getcwd(), "out"), ignore_errors=True)
249
+
250
+ command = [
251
+ "python", "-m", "apps.eval", "--results_path", "out", "--name", "out",
252
+ "--dataroot", f'{os.getcwd()}/AnchorUDF/pics/my_test', "--test_folder_path", "1",
253
+ "--load_netG_checkpoint_path", os.path.join(os.getcwd(), 'netG_epoch_59'),
254
+ "--anchor", "--num_steps", "2", "--filter_val", "0.0075",
255
+ "--max_dist", "0.05", "--num_gen_points", "850000", "--gamma", "1",
256
+ "--aug_bri", "1.5", "--aug_con", "0.5", "--aug_sat", "0.5", "--aug_hue", "0.5"
257
+ ]
258
+
259
+ subprocess.run(command, cwd=os.path.join(os.getcwd(), "AnchorUDF"))
260
+ print('garment generated!!!')
261
+
262
+ @app.route('/process', methods=['POST'])
263
+ def process():
264
+ data = request.json
265
+ filename = data.get('filename')
266
+
267
+ if not filename:
268
+ return jsonify({"error": "Filename is required"}), 400
269
+
270
+ processed_image_path = process_image(filename)
271
+ mask_path = generate_mask(processed_image_path)
272
+ run_garment_inference()
273
+
274
+ try:
275
+ print('generating ply...')
276
+ cleaned_ply_path = process_mesh(scale_factor=0.5, nb_neighbors=40, std_ratio=1.0, voxel_size=0.003)
277
+ print('ply generated')
278
+ except FileNotFoundError as e:
279
+ import traceback
280
+ traceback.print_exc()
281
+ return jsonify({"error": str(e)}), 500
282
+
283
+ return send_file(cleaned_ply_path, as_attachment=True)
284
+
285
+
286
+ # Run the Flask app
287
+ if __name__ == "__main__":
288
+ app.run(host='0.0.0.0', port=6100)