Update app.py
Browse files
app.py
CHANGED
@@ -31,7 +31,7 @@ BATCH_SIZE = 4 # Process 4 frames at a time
|
|
31 |
|
32 |
# Load model
|
33 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
34 |
-
model = attempt_load("
|
35 |
model.to(device).eval() # Move model to device and set to evaluation mode
|
36 |
|
37 |
# Function to process video and detect ball
|
@@ -114,10 +114,10 @@ def process_video(video_path):
|
|
114 |
def poly_func(x, a, b, c):
|
115 |
return a * x**2 + b * x + c
|
116 |
|
117 |
-
# Predict trajectory and
|
118 |
def predict_trajectory(positions, frame_numbers, frame_width, frame_height):
|
119 |
if len(positions) < 3:
|
120 |
-
return None, "Insufficient detections for trajectory prediction"
|
121 |
|
122 |
x_coords = [p[0] for p in positions]
|
123 |
y_coords = [p[1] for p in positions]
|
@@ -128,7 +128,7 @@ def predict_trajectory(positions, frame_numbers, frame_width, frame_height):
|
|
128 |
popt_x, _ = curve_fit(poly_func, frames, x_coords)
|
129 |
popt_y, _ = curve_fit(poly_func, frames, y_coords)
|
130 |
except:
|
131 |
-
return None, "Failed to fit trajectory"
|
132 |
|
133 |
# Extrapolate to stumps
|
134 |
frame_max = max(frames) + 10
|
@@ -136,9 +136,13 @@ def predict_trajectory(positions, frame_numbers, frame_width, frame_height):
|
|
136 |
x_pred = poly_func(future_frames, *popt_x)
|
137 |
y_pred = poly_func(future_frames, *popt_y)
|
138 |
|
139 |
-
#
|
140 |
stump_x = frame_width / 2
|
141 |
stump_y = frame_height
|
|
|
|
|
|
|
|
|
142 |
stump_hit = False
|
143 |
for x, y in zip(x_pred, y_pred):
|
144 |
if abs(y - stump_y) < 50 and abs(x - stump_x) < STUMP_WIDTH * frame_width / PITCH_WIDTH:
|
@@ -146,7 +150,7 @@ def predict_trajectory(positions, frame_numbers, frame_width, frame_height):
|
|
146 |
break
|
147 |
|
148 |
lbw_decision = "OUT" if stump_hit else "NOT OUT"
|
149 |
-
return list(zip(future_frames, x_pred, y_pred)), lbw_decision
|
150 |
|
151 |
# Map pitch location
|
152 |
def map_pitch(bounce_point, frame_width, frame_height):
|
@@ -177,27 +181,7 @@ def estimate_speed(positions, frame_numbers, frame_rate, frame_width):
|
|
177 |
avg_speed_kmh = np.mean(speeds) * 3.6
|
178 |
return avg_speed_kmh, "Speed calculated successfully"
|
179 |
|
180 |
-
#
|
181 |
-
def create_pitch_map(pitch_x, pitch_y):
|
182 |
-
fig = go.Figure()
|
183 |
-
fig.add_shape(
|
184 |
-
type="rect", x0=-PITCH_WIDTH/2, y0=0, x1=PITCH_WIDTH/2, y1=PITCH_LENGTH,
|
185 |
-
line=dict(color="Green"), fillcolor="Green", opacity=0.3
|
186 |
-
)
|
187 |
-
fig.add_shape(
|
188 |
-
type="rect", x0=-STUMP_WIDTH/2, y0=PITCH_LENGTH-0.1, x1=STUMP_WIDTH/2, y1=PITCH_LENGTH,
|
189 |
-
line=dict(color="Brown"), fillcolor="Brown"
|
190 |
-
)
|
191 |
-
if pitch_x is not None and pitch_y is not None:
|
192 |
-
fig.add_trace(go.Scatter(x=[pitch_x], y=[pitch_y], mode="markers", marker=dict(size=10, color="Red"), name="Bounce Point"))
|
193 |
-
|
194 |
-
fig.update_layout(
|
195 |
-
title="Pitch Map", xaxis_title="Width (m)", yaxis_title="Length (m)",
|
196 |
-
xaxis_range=[-PITCH_WIDTH/2, PITCH_WIDTH/2], yaxis_range=[0, PITCH_LENGTH]
|
197 |
-
)
|
198 |
-
return fig
|
199 |
-
|
200 |
-
# Main Gradio function
|
201 |
def drs_analysis(video):
|
202 |
# Video is a file path (string) in Hugging Face Spaces
|
203 |
video_path = video if isinstance(video, str) else "temp_video.mp4"
|
@@ -205,27 +189,80 @@ def drs_analysis(video):
|
|
205 |
with open(video_path, "wb") as f:
|
206 |
f.write(video.read())
|
207 |
|
|
|
208 |
positions, frame_numbers, bounce_point, frame_rate, frame_width, frame_height = process_video(video_path)
|
209 |
if not positions:
|
210 |
return None, None, "No ball detected in video", None
|
211 |
|
212 |
-
|
|
|
213 |
if trajectory is None:
|
214 |
return None, None, lbw_decision, None
|
215 |
|
216 |
pitch_x, pitch_y = map_pitch(bounce_point, frame_width, frame_height)
|
217 |
speed_kmh, speed_status = estimate_speed(positions, frame_numbers, frame_rate, frame_width)
|
218 |
|
219 |
-
|
220 |
-
|
221 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
222 |
|
223 |
-
|
|
|
224 |
|
225 |
if not isinstance(video, str):
|
226 |
os.remove(video_path)
|
227 |
|
228 |
-
return
|
229 |
|
230 |
# Gradio interface
|
231 |
with gr.Blocks() as demo:
|
|
|
31 |
|
32 |
# Load model
|
33 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
34 |
+
model = attempt_load("best.pt") # Load without map_location
|
35 |
model.to(device).eval() # Move model to device and set to evaluation mode
|
36 |
|
37 |
# Function to process video and detect ball
|
|
|
114 |
def poly_func(x, a, b, c):
|
115 |
return a * x**2 + b * x + c
|
116 |
|
117 |
+
# Predict trajectory and wicket inline path
|
118 |
def predict_trajectory(positions, frame_numbers, frame_width, frame_height):
|
119 |
if len(positions) < 3:
|
120 |
+
return None, None, "Insufficient detections for trajectory prediction"
|
121 |
|
122 |
x_coords = [p[0] for p in positions]
|
123 |
y_coords = [p[1] for p in positions]
|
|
|
128 |
popt_x, _ = curve_fit(poly_func, frames, x_coords)
|
129 |
popt_y, _ = curve_fit(poly_func, frames, y_coords)
|
130 |
except:
|
131 |
+
return None, None, "Failed to fit trajectory"
|
132 |
|
133 |
# Extrapolate to stumps
|
134 |
frame_max = max(frames) + 10
|
|
|
136 |
x_pred = poly_func(future_frames, *popt_x)
|
137 |
y_pred = poly_func(future_frames, *popt_y)
|
138 |
|
139 |
+
# Wicket inline path (center line toward stumps)
|
140 |
stump_x = frame_width / 2
|
141 |
stump_y = frame_height
|
142 |
+
inline_x = np.linspace(min(x_coords), stump_x, 100)
|
143 |
+
inline_y = np.interp(inline_x, x_pred, y_pred)
|
144 |
+
|
145 |
+
# Check if trajectory hits stumps
|
146 |
stump_hit = False
|
147 |
for x, y in zip(x_pred, y_pred):
|
148 |
if abs(y - stump_y) < 50 and abs(x - stump_x) < STUMP_WIDTH * frame_width / PITCH_WIDTH:
|
|
|
150 |
break
|
151 |
|
152 |
lbw_decision = "OUT" if stump_hit else "NOT OUT"
|
153 |
+
return list(zip(future_frames, x_pred, y_pred)), list(zip(inline_x, inline_y)), lbw_decision
|
154 |
|
155 |
# Map pitch location
|
156 |
def map_pitch(bounce_point, frame_width, frame_height):
|
|
|
181 |
avg_speed_kmh = np.mean(speeds) * 3.6
|
182 |
return avg_speed_kmh, "Speed calculated successfully"
|
183 |
|
184 |
+
# Main Gradio function with video overlay
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
def drs_analysis(video):
|
186 |
# Video is a file path (string) in Hugging Face Spaces
|
187 |
video_path = video if isinstance(video, str) else "temp_video.mp4"
|
|
|
189 |
with open(video_path, "wb") as f:
|
190 |
f.write(video.read())
|
191 |
|
192 |
+
# Process video for detections
|
193 |
positions, frame_numbers, bounce_point, frame_rate, frame_width, frame_height = process_video(video_path)
|
194 |
if not positions:
|
195 |
return None, None, "No ball detected in video", None
|
196 |
|
197 |
+
# Predict trajectory and wicket path
|
198 |
+
trajectory, inline_path, lbw_decision = predict_trajectory(positions, frame_numbers, frame_width, frame_height)
|
199 |
if trajectory is None:
|
200 |
return None, None, lbw_decision, None
|
201 |
|
202 |
pitch_x, pitch_y = map_pitch(bounce_point, frame_width, frame_height)
|
203 |
speed_kmh, speed_status = estimate_speed(positions, frame_numbers, frame_rate, frame_width)
|
204 |
|
205 |
+
# Create output video with overlays
|
206 |
+
output_path = "output_video.mp4"
|
207 |
+
cap = cv2.VideoCapture(video_path)
|
208 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
209 |
+
out = cv2.VideoWriter(output_path, fourcc, frame_rate, (frame_width, frame_height))
|
210 |
+
|
211 |
+
frame_count = 0
|
212 |
+
positions_dict = dict(zip(frame_numbers, positions))
|
213 |
+
|
214 |
+
while cap.isOpened():
|
215 |
+
ret, frame = cap.read()
|
216 |
+
if not ret:
|
217 |
+
break
|
218 |
+
|
219 |
+
# Skip frames for consistency with detection
|
220 |
+
if frame_count % FRAME_SKIP != 0:
|
221 |
+
frame_count += 1
|
222 |
+
continue
|
223 |
+
|
224 |
+
# Overlay ball trajectory (red) and wicket inline path (blue)
|
225 |
+
if frame_count in positions_dict:
|
226 |
+
cv2.circle(frame, (int(positions_dict[frame_count][0]), int(positions_dict[frame_count][1])), 5, (0, 0, 255), -1) # Red dot
|
227 |
+
if trajectory:
|
228 |
+
traj_x = [int(t[1]) for t in trajectory if t[0] >= frame_count]
|
229 |
+
traj_y = [int(t[2]) for t in trajectory if t[0] >= frame_count]
|
230 |
+
if traj_x and traj_y:
|
231 |
+
for i in range(1, len(traj_x)):
|
232 |
+
cv2.line(frame, (traj_x[i-1], traj_y[i-1]), (traj_x[i], traj_y[i]), (0, 0, 255), 2) # Red line
|
233 |
+
if inline_path:
|
234 |
+
inline_x = [int(x) for x, _ in inline_path]
|
235 |
+
inline_y = [int(y) for _, y in inline_path]
|
236 |
+
if inline_x and inline_y:
|
237 |
+
for i in range(1, len(inline_x)):
|
238 |
+
cv2.line(frame, (inline_x[i-1], inline_y[i-1]), (inline_x[i], inline_y[i]), (255, 0, 0), 2) # Blue line
|
239 |
+
|
240 |
+
# Overlay pitch map in top-right corner
|
241 |
+
if pitch_x is not None and pitch_y is not None:
|
242 |
+
map_width = 200
|
243 |
+
map_height = int(map_width * PITCH_LENGTH / PITCH_WIDTH)
|
244 |
+
pitch_map = np.zeros((map_height, map_width, 3), dtype=np.uint8)
|
245 |
+
pitch_map[:] = (0, 255, 0) # Green pitch
|
246 |
+
cv2.rectangle(pitch_map, (0, map_height-10), (map_width, map_height), (0, 51, 51), -1) # Brown stumps
|
247 |
+
bounce_x = int((pitch_x + PITCH_WIDTH/2) / PITCH_WIDTH * map_width)
|
248 |
+
bounce_y = int((1 - pitch_y / PITCH_LENGTH) * map_height)
|
249 |
+
cv2.circle(pitch_map, (bounce_x, bounce_y), 5, (0, 0, 255), -1) # Red bounce point
|
250 |
+
frame[0:map_height, frame_width-map_width:frame_width] = cv2.resize(pitch_map, (map_width, map_height))
|
251 |
+
|
252 |
+
# Add text annotations
|
253 |
+
text = f"LBW: {lbw_decision}\nSpeed: {speed_kmh:.2f} km/h"
|
254 |
+
cv2.putText(frame, text, (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA)
|
255 |
+
|
256 |
+
out.write(frame)
|
257 |
+
frame_count += 1
|
258 |
|
259 |
+
cap.release()
|
260 |
+
out.release()
|
261 |
|
262 |
if not isinstance(video, str):
|
263 |
os.remove(video_path)
|
264 |
|
265 |
+
return None, None, None, output_path
|
266 |
|
267 |
# Gradio interface
|
268 |
with gr.Blocks() as demo:
|