File size: 14,096 Bytes
b4b1c15
8166792
 
 
297af59
8166792
b4b1c15
661e202
00ab2e7
3ece440
 
 
8166792
 
 
661e202
8166792
6ea8a1f
f796a73
512f124
 
a45b657
297af59
 
8569624
a45b657
8569624
 
297af59
 
8569624
 
 
 
 
 
 
 
 
 
a45b657
8569624
 
7a68aca
6ea8a1f
 
f796a73
8166792
 
 
b4b1c15
 
2f27314
a45b657
b4b1c15
a45b657
 
 
 
 
 
 
 
8569624
2f27314
7a68aca
661e202
 
 
 
 
8166792
8569624
8166792
a45b657
 
 
 
 
 
 
 
 
 
b4b1c15
8166792
7a68aca
a45b657
8166792
 
 
 
 
 
 
 
7a68aca
661e202
7a68aca
2f27314
b4b1c15
 
 
 
 
 
 
 
2f27314
b4b1c15
 
661e202
831d4de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a9bcd01
831d4de
 
a9bcd01
 
 
 
 
3ece440
a9bcd01
 
 
 
 
 
 
 
 
 
3ece440
 
 
 
 
 
a4031b7
831d4de
3ece440
 
 
 
 
 
 
a9bcd01
 
831d4de
a4031b7
a9bcd01
3ece440
a4031b7
da90e52
831d4de
 
a4031b7
a9bcd01
 
 
 
3ece440
 
a4031b7
da90e52
3ece440
 
 
a4031b7
831d4de
a4031b7
831d4de
a4031b7
 
 
a9bcd01
 
831d4de
8166792
 
 
 
c5c62a8
b4b1c15
 
 
 
 
 
 
 
 
2f27314
b4b1c15
 
 
 
 
 
 
 
8166792
b4b1c15
8166792
00ab2e7
 
 
8166792
 
 
 
 
 
661e202
 
 
 
 
 
8166792
 
 
 
 
 
 
 
661e202
 
 
00ab2e7
8166792
b4b1c15
8166792
00ab2e7
 
 
 
 
8166792
00ab2e7
8166792
 
 
 
 
 
 
 
661e202
 
 
 
 
00ab2e7
661e202
8166792
 
 
 
 
 
 
 
00ab2e7
 
661e202
 
 
8166792
b4b1c15
8166792
00ab2e7
 
 
 
8166792
b4b1c15
 
8166792
b4b1c15
ef0cbed
 
 
 
 
 
 
 
 
 
 
 
8166792
 
00ab2e7
8166792
 
661e202
8166792
 
661e202
00ab2e7
8166792
 
6377841
 
26ed140
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
from ultralytics import YOLO
import cv2
import gradio as gr
import numpy as np
import spaces
import os
import torch
import utils
import plotly.graph_objects as go
from io import BytesIO
from PIL import Image
import base64

from image_segmenter import ImageSegmenter
from monocular_depth_estimator import MonocularDepthEstimator
from point_cloud_generator import display_pcd






device = torch.device("cpu")  # Start in CPU mode

def initialize_gpu():
    """Ensure ZeroGPU assigns a GPU before initializing CUDA"""
    global device
    try:
        with spaces.GPU():  # Ensures ZeroGPU assigns a GPU
            torch.cuda.empty_cache()  # Prevent leftover memory issues
            if torch.cuda.is_available():
                device = torch.device("cuda")
                print(f"βœ… GPU initialized: {torch.cuda.get_device_name(0)}")
            else:
                print("❌ No GPU detected after ZeroGPU allocation.")
                device = torch.device("cpu")
    except Exception as e:
        print(f"🚨 GPU initialization failed: {e}")
        device = torch.device("cpu")


# Run GPU initialization before using CUDA
initialize_gpu()




# params
CANCEL_PROCESSING = False

img_seg = ImageSegmenter(model_type="yolov8s-seg")
depth_estimator = MonocularDepthEstimator(model_type="midas_v21_small_256")

@spaces.GPU  # Ensures ZeroGPU assigns a GPU
def process_image(image):
    image = utils.resize(image)
    image_segmentation, objects_data = img_seg.predict(image)
    depthmap, depth_colormap = depth_estimator.make_prediction(image)
    dist_image = utils.draw_depth_info(image, depthmap, objects_data)
    objs_pcd = utils.generate_obj_pcd(depthmap, objects_data)
    plot_fig = display_pcd(objs_pcd)
    return image_segmentation, depth_colormap, dist_image, plot_fig



@spaces.GPU  # Requests GPU for depth estimation
def test_process_img(image):
    image = utils.resize(image)
    image_segmentation, objects_data = img_seg.predict(image)
    depthmap, depth_colormap = depth_estimator.make_prediction(image)
    return image_segmentation, objects_data, depthmap, depth_colormap

@spaces.GPU
def process_video(vid_path=None):
    vid_cap = cv2.VideoCapture(vid_path)
    while vid_cap.isOpened():
        ret, frame = vid_cap.read()
        if ret:
            print("making predictions ....")
            frame = utils.resize(frame)
            image_segmentation, objects_data = img_seg.predict(frame)
            depthmap, depth_colormap = depth_estimator.make_prediction(frame)
            dist_image = utils.draw_depth_info(frame, depthmap, objects_data)
            yield cv2.cvtColor(image_segmentation, cv2.COLOR_BGR2RGB), depth_colormap, cv2.cvtColor(dist_image, cv2.COLOR_BGR2RGB)
    return None



def update_segmentation_options(options):
    img_seg.is_show_bounding_boxes = True if 'Show Boundary Box' in options else False
    img_seg.is_show_segmentation = True if 'Show Segmentation Region' in options else False
    img_seg.is_show_segmentation_boundary = True if 'Show Segmentation Boundary' in options else False

def update_confidence_threshold(thres_val):
    img_seg.confidence_threshold = thres_val/100

@spaces.GPU  # Ensures YOLO + MiDaS get GPU access
def model_selector(model_type):
    global img_seg, depth_estimator

    if "Small - Better performance and less accuracy" == model_type:
        midas_model, yolo_model = "midas_v21_small_256", "yolov8s-seg"
    elif "Medium - Balanced performance and accuracy" == model_type:
        midas_model, yolo_model = "dpt_hybrid_384", "yolov8m-seg"
    elif "Large - Slow performance and high accuracy" == model_type:
        midas_model, yolo_model = "dpt_large_384", "yolov8l-seg"
    else:
        midas_model, yolo_model = "midas_v21_small_256", "yolov8s-seg"

    img_seg = ImageSegmenter(model_type=yolo_model)
    depth_estimator = MonocularDepthEstimator(model_type=midas_model)

    # START 
    # added for lens studio 

    
def get_box_vertices(bbox):
    """Convert bbox to corner vertices"""
    x1, y1, x2, y2 = bbox
    return [
        [x1, y1],  # top-left
        [x2, y1],  # top-right
        [x2, y2],  # bottom-right
        [x1, y2]   # bottom-left
    ]

def depth_at_center(depth_map, bbox):
    """Get depth at center of bounding box"""
    x1, y1, x2, y2 = bbox
    center_x = int((x1 + x2) / 2)
    center_y = int((y1 + y2) / 2)
    
    # Sample a small region around center for stability
    region = depth_map[
        max(0, center_y-2):min(depth_map.shape[0], center_y+3),
        max(0, center_x-2):min(depth_map.shape[1], center_x+3)
    ]
    return np.median(region)

def get_camera_matrix(depth_estimator):
    """Get camera calibration matrix"""
    return {
        "fx": depth_estimator.fx_depth,
        "fy": depth_estimator.fy_depth,
        "cx": depth_estimator.cx_depth,
        "cy": depth_estimator.cy_depth
    }


@spaces.GPU
def get_detection_data(image):
    """Get structured detection data with depth information, using Base64 image encoding."""

    def decode_base64_image(base64_string):
        """Decodes Base64 string into a NumPy image."""
        try:
            print(f"πŸ” Received Base64 String (Truncated): {base64_string[:50]}...")  # Debugging
            img_data = base64.b64decode(base64_string)
            img = Image.open(BytesIO(img_data))
            img = np.array(img)
            return cv2.cvtColor(img, cv2.COLOR_RGB2BGR)  # Convert to BGR for OpenCV
        except Exception as e:
            print(f"🚨 Error decoding base64 image: {e}")
            return None

    def encode_base64_image(image):
        """Encodes a NumPy image into a Base64 string."""
        try:
            _, buffer = cv2.imencode('.png', image)
            return base64.b64encode(buffer).decode("utf-8")
        except Exception as e:
            print(f"🚨 Error encoding image to Base64: {e}")
            return None

    try:
        if not isinstance(image, str):
            print("🚨 Error: Expected Base64 string but received:", type(image))
            return {"error": "Invalid input format. Expected Base64-encoded image."}

        image = decode_base64_image(image)
        if image is None:
            return {"error": "Base64 decoding failed. Ensure correct encoding."}

        # Resize image
        image = utils.resize(image)

        # Extract dimensions
        height, width = image.shape[:2]

        # Get detections and depth
        image_segmentation, objects_data = img_seg.predict(image)
        depthmap, depth_colormap = depth_estimator.make_prediction(image)

        # Encode results as Base64
        segmentation_b64 = encode_base64_image(image_segmentation)
        depth_b64 = encode_base64_image(depth_colormap)

        if segmentation_b64 is None or depth_b64 is None:
            return {"error": "Failed to encode output images."}

        return {
            "detections": objects_data,  # Keeping as original
            "depth_map": depth_b64,
            "segmentation": segmentation_b64,
            "image_size": {"width": width, "height": height},
        }

    except Exception as e:
        print(f"🚨 Error in get_detection_data: {str(e)}")
        return {"error": str(e)}

     


def cancel():
    CANCEL_PROCESSING = True

if __name__ == "__main__":

    # testing
    # img_1 = cv2.imread("assets/images/bus.jpg")
    # img_1 = utils.resize(img_1)

    # image_segmentation, objects_data, depthmap, depth_colormap = test_process_img(img_1)
    # final_image = utils.draw_depth_info(image_segmentation, depthmap, objects_data)
    # objs_pcd = utils.generate_obj_pcd(depthmap, objects_data)
    # # print(objs_pcd[0][0])
    # display_pcd(objs_pcd, use_matplotlib=True)

    # cv2.imshow("Segmentation", image_segmentation)
    # cv2.imshow("Depth", depthmap*objects_data[2][3])
    # cv2.imshow("Final", final_image)

    # cv2.waitKey(0)
    # cv2.destroyAllWindows()

    # gradio gui app
    with gr.Blocks() as my_app:

        # title
        gr.Markdown("<h1><center>Simultaneous Segmentation and Depth Estimation</center></h1>")
        gr.Markdown("<h3><center>Created by Vaishanth</center></h3>")
        gr.Markdown("<h3><center>This model estimates the depth of segmented objects.</center></h3>")

        # tabs
        with gr.Tab("Image"):
            with gr.Row():
                with gr.Column(scale=1):
                    img_input = gr.Image()
                    model_type_img = gr.Dropdown(
                        ["Small - Better performance and less accuracy", 
                         "Medium - Balanced performance and accuracy", 
                         "Large - Slow performance and high accuracy"], 
                        label="Model Type", value="Small - Better performance and less accuracy",
                        info="Select the inference model before running predictions!")
                    options_checkbox_img = gr.CheckboxGroup(["Show Boundary Box", "Show Segmentation Region", "Show Segmentation Boundary"], label="Options")
                    conf_thres_img = gr.Slider(1, 100, value=60, label="Confidence Threshold", info="Choose the threshold above which objects should be detected")
                    submit_btn_img = gr.Button(value="Predict")                    

                with gr.Column(scale=2):
                    with gr.Row():
                        segmentation_img_output = gr.Image(height=300, label="Segmentation")
                        depth_img_output = gr.Image(height=300, label="Depth Estimation")
                    
                    with gr.Row():
                        dist_img_output = gr.Image(height=300, label="Distance")
                        pcd_img_output = gr.Plot(label="Point Cloud")
            
            gr.Markdown("## Sample Images")
            gr.Examples(
                examples=[os.path.join(os.path.dirname(__file__), "assets/images/baggage_claim.jpg"),
                          os.path.join(os.path.dirname(__file__), "assets/images/kitchen_2.png"),
                          os.path.join(os.path.dirname(__file__), "assets/images/soccer.jpg"),
                          os.path.join(os.path.dirname(__file__), "assets/images/room_2.png"),
                          os.path.join(os.path.dirname(__file__), "assets/images/living_room.jpg")],
                inputs=img_input,
                outputs=[segmentation_img_output, depth_img_output, dist_img_output, pcd_img_output],
                fn=process_image,
                cache_examples=True,
            )

        with gr.Tab("Video"):
            with gr.Row():
                with gr.Column(scale=1):
                    vid_input = gr.Video()
                    model_type_vid = gr.Dropdown(
                        ["Small - Better performance and less accuracy", 
                         "Medium - Balanced performance and accuracy", 
                         "Large - Slow performance and high accuracy"], 
                        label="Model Type", value="Small - Better performance and less accuracy",
                        info="Select the inference model before running predictions!")
                    
                    options_checkbox_vid = gr.CheckboxGroup(["Show Boundary Box", "Show Segmentation Region", "Show Segmentation Boundary"], label="Options")
                    conf_thres_vid = gr.Slider(1, 100, value=60, label="Confidence Threshold", info="Choose the threshold above which objects should be detected")
                    with gr.Row():
                        cancel_btn = gr.Button(value="Cancel")
                        submit_btn_vid = gr.Button(value="Predict")
            
                with gr.Column(scale=2):
                    with gr.Row():
                        segmentation_vid_output = gr.Image(height=300, label="Segmentation")
                        depth_vid_output = gr.Image(height=300, label="Depth Estimation")
                    
                    with gr.Row():
                        dist_vid_output = gr.Image(height=300, label="Distance")
            
            gr.Markdown("## Sample Videos")
            gr.Examples(
                examples=[os.path.join(os.path.dirname(__file__), "assets/videos/input_video.mp4"),
                          os.path.join(os.path.dirname(__file__), "assets/videos/driving.mp4"),
                          os.path.join(os.path.dirname(__file__), "assets/videos/overpass.mp4"),
                          os.path.join(os.path.dirname(__file__), "assets/videos/walking.mp4")],
                inputs=vid_input,
                # outputs=vid_output,
                # fn=vid_segmenation,
            )
            
        # Add a new hidden tab or interface for the API endpoint
        with gr.Tab("API", visible=False):  # Hidden from UI but accessible via API
            input_image = gr.Image()
            output_json = gr.JSON()
            gr.Interface(
                fn=get_detection_data,
                inputs=input_image,
                outputs=output_json,
                title="Get Detection Data",
                api_name="get_detection_data"  # This sets the endpoint name
            )
            

        # image tab logic
        submit_btn_img.click(process_image, inputs=img_input, outputs=[segmentation_img_output, depth_img_output, dist_img_output, pcd_img_output])
        options_checkbox_img.change(update_segmentation_options, options_checkbox_img, [])
        conf_thres_img.change(update_confidence_threshold, conf_thres_img, [])
        model_type_img.change(model_selector, model_type_img, [])

        # video tab logic
        submit_btn_vid.click(process_video, inputs=vid_input, outputs=[segmentation_vid_output, depth_vid_output, dist_vid_output])
        model_type_vid.change(model_selector, model_type_vid, [])
        cancel_btn.click(cancel, inputs=[], outputs=[])
        options_checkbox_vid.change(update_segmentation_options, options_checkbox_vid, [])
        conf_thres_vid.change(update_confidence_threshold, conf_thres_vid, [])      

    my_app.queue(max_size=20).launch(share=True)  # Add share=True here