Spaces:
Runtime error
Runtime error
initial commit
Browse files- README.md +6 -4
- app.py +170 -0
- requirements.txt +7 -0
README.md
CHANGED
|
@@ -1,12 +1,14 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
colorFrom: purple
|
| 5 |
-
colorTo:
|
| 6 |
sdk: streamlit
|
| 7 |
sdk_version: 1.10.0
|
|
|
|
| 8 |
app_file: app.py
|
| 9 |
-
|
|
|
|
| 10 |
license: mit
|
| 11 |
---
|
| 12 |
|
|
|
|
| 1 |
---
|
| 2 |
+
title: YOLOS Demo (Balloons)
|
| 3 |
+
emoji: π
|
| 4 |
colorFrom: purple
|
| 5 |
+
colorTo: red
|
| 6 |
sdk: streamlit
|
| 7 |
sdk_version: 1.10.0
|
| 8 |
+
python_version: 3.9.13
|
| 9 |
app_file: app.py
|
| 10 |
+
models: zoheb/yolos-small-balloon
|
| 11 |
+
pinned: true
|
| 12 |
license: mit
|
| 13 |
---
|
| 14 |
|
app.py
ADDED
|
@@ -0,0 +1,170 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import shutil
|
| 2 |
+
import cv2
|
| 3 |
+
from PIL import Image
|
| 4 |
+
import streamlit as st
|
| 5 |
+
from transformers import AutoModelForObjectDetection, AutoFeatureExtractor
|
| 6 |
+
import torch
|
| 7 |
+
import matplotlib.pyplot as plt
|
| 8 |
+
from stqdm import stqdm
|
| 9 |
+
from pathlib import Path
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
# Load the model
|
| 13 |
+
best_model_path = "zoheb/yolos-small-balloon"
|
| 14 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 15 |
+
feature_extractor = AutoFeatureExtractor.from_pretrained(best_model_path, size=512, max_size=864)
|
| 16 |
+
model_pt = AutoModelForObjectDetection.from_pretrained(best_model_path).to(device)
|
| 17 |
+
|
| 18 |
+
# colors for visualization
|
| 19 |
+
COLORS = [[0.000, 0.447, 0.741], [0.850, 0.325, 0.098], [0.929, 0.694, 0.125],
|
| 20 |
+
[0.494, 0.184, 0.556], [0.466, 0.674, 0.188], [0.301, 0.745, 0.933]]
|
| 21 |
+
|
| 22 |
+
# Convert Video to Frames
|
| 23 |
+
def video_to_frames(video, dir):
|
| 24 |
+
cap = cv2.VideoCapture(str(video))
|
| 25 |
+
success, image = cap.read()
|
| 26 |
+
|
| 27 |
+
frame_count = 0
|
| 28 |
+
while success:
|
| 29 |
+
frameId = int(round(cap.get(1))) # current frame number
|
| 30 |
+
if frameId % 5 == 0:
|
| 31 |
+
cv2.imwrite(f"{str(dir)}/frame_{frame_count}.jpg", image)
|
| 32 |
+
frame_count += 1
|
| 33 |
+
success, image = cap.read()
|
| 34 |
+
|
| 35 |
+
cap.release()
|
| 36 |
+
#print (f"No. of frames {frame_count}")
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
# for output bounding box post-processing
|
| 40 |
+
def box_cxcywh_to_xyxy(x):
|
| 41 |
+
x_c, y_c, w, h = x.unbind(1)
|
| 42 |
+
b = [(x_c - 0.5 * w), (y_c - 0.5 * h),
|
| 43 |
+
(x_c + 0.5 * w), (y_c + 0.5 * h)]
|
| 44 |
+
return torch.stack(b, dim=1)
|
| 45 |
+
|
| 46 |
+
# rescale bboxes
|
| 47 |
+
def rescale_bboxes(out_bbox, size):
|
| 48 |
+
img_w, img_h = size
|
| 49 |
+
b = box_cxcywh_to_xyxy(out_bbox)
|
| 50 |
+
b = b * torch.tensor([img_w, img_h, img_w, img_h], dtype=torch.float32)
|
| 51 |
+
return b
|
| 52 |
+
|
| 53 |
+
# Save predicted frame
|
| 54 |
+
def save_results(pil_img, prob, boxes, mod_img_path):
|
| 55 |
+
plt.figure(figsize=(18,10))
|
| 56 |
+
plt.imshow(pil_img)
|
| 57 |
+
id2label = {0: 'balloon'}
|
| 58 |
+
ax = plt.gca()
|
| 59 |
+
colors = COLORS * 100
|
| 60 |
+
for p, (xmin, ymin, xmax, ymax), c in zip(prob, boxes.tolist(), colors):
|
| 61 |
+
ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin,
|
| 62 |
+
fill=False, color=c, linewidth=3))
|
| 63 |
+
cl = p.argmax()
|
| 64 |
+
text = f'{id2label[cl.item()]}: {p[cl]:0.2f}'
|
| 65 |
+
ax.text(xmin, ymin, text, fontsize=15,
|
| 66 |
+
bbox=dict(facecolor='yellow', alpha=0.5))
|
| 67 |
+
plt.axis('off')
|
| 68 |
+
plt.tight_layout(pad=0)
|
| 69 |
+
plt.savefig(mod_img_path, transparent=True)
|
| 70 |
+
plt.close()
|
| 71 |
+
|
| 72 |
+
# Save predictions
|
| 73 |
+
def save_predictions(image, outputs, mod_img_path, threshold=0.9):
|
| 74 |
+
# keep only predictions with confidence >= threshold
|
| 75 |
+
probas = outputs.logits.softmax(-1)[0, :, :-1]
|
| 76 |
+
keep = probas.max(-1).values > threshold
|
| 77 |
+
|
| 78 |
+
# convert predicted boxes from [0; 1] to image scales
|
| 79 |
+
bboxes_scaled = rescale_bboxes(outputs.pred_boxes[0, keep].cpu(), image.size)
|
| 80 |
+
|
| 81 |
+
# save results
|
| 82 |
+
save_results(image, probas[keep], bboxes_scaled, mod_img_path)
|
| 83 |
+
|
| 84 |
+
# Predict on frames
|
| 85 |
+
def predict_on_frames(dir, mod_dir):
|
| 86 |
+
files = [f for f in dir.glob('*.jpg') if f.is_file()]
|
| 87 |
+
#for sorting the file names properly
|
| 88 |
+
files.sort(key = lambda x: int(x.stem[6:]))
|
| 89 |
+
|
| 90 |
+
for i in stqdm(range(len(files)), desc="Generating... this is a slow task"):
|
| 91 |
+
filename = Path(dir, files[i])
|
| 92 |
+
#print(filename)
|
| 93 |
+
#reading each files
|
| 94 |
+
img = Image.open(str(filename))
|
| 95 |
+
# extract features
|
| 96 |
+
img_ftr = feature_extractor(images=img, return_tensors="pt")
|
| 97 |
+
pixel_values = img_ftr["pixel_values"].to(device)
|
| 98 |
+
# forward pass to get class logits and bounding boxes
|
| 99 |
+
outputs = model_pt(pixel_values=pixel_values)
|
| 100 |
+
mod_img_path = Path(mod_dir, files[i].name)
|
| 101 |
+
save_predictions(img, outputs, mod_img_path)
|
| 102 |
+
|
| 103 |
+
# Convert frames to video
|
| 104 |
+
def frames_to_video(dir, path, fps=5):
|
| 105 |
+
frame_array = []
|
| 106 |
+
files = [f for f in dir.glob('*.jpg') if f.is_file()]
|
| 107 |
+
#for sorting the file names properly
|
| 108 |
+
files.sort(key = lambda x: int(x.stem[6:]))
|
| 109 |
+
for file in files:
|
| 110 |
+
filename = Path(dir, file)
|
| 111 |
+
#reading each files
|
| 112 |
+
img = cv2.imread(str(filename))
|
| 113 |
+
height, width, _ = img.shape
|
| 114 |
+
size = (width, height)
|
| 115 |
+
#print(filename)
|
| 116 |
+
#inserting the frames into an image array
|
| 117 |
+
frame_array.append(img)
|
| 118 |
+
out = cv2.VideoWriter(str(path), cv2.VideoWriter_fourcc(*'DIVX'), fps, size)
|
| 119 |
+
for item in frame_array:
|
| 120 |
+
# writing to a image array
|
| 121 |
+
out.write(item)
|
| 122 |
+
out.release()
|
| 123 |
+
|
| 124 |
+
|
| 125 |
+
# Main
|
| 126 |
+
if __name__=='__main__':
|
| 127 |
+
st.title('Detect Balloons using YOLOS')
|
| 128 |
+
|
| 129 |
+
# All dir and Files
|
| 130 |
+
BASE_DIR = Path(__file__).parent.absolute()
|
| 131 |
+
|
| 132 |
+
FRAMES_DIR = Path(BASE_DIR, "extracted_images")
|
| 133 |
+
MOD_DIR = Path(BASE_DIR, "modified_images")
|
| 134 |
+
|
| 135 |
+
if FRAMES_DIR.exists() and FRAMES_DIR.is_dir():
|
| 136 |
+
shutil.rmtree(FRAMES_DIR)
|
| 137 |
+
FRAMES_DIR.mkdir(parents=True, exist_ok=True)
|
| 138 |
+
|
| 139 |
+
if MOD_DIR.exists() and MOD_DIR.is_dir():
|
| 140 |
+
shutil.rmtree(MOD_DIR)
|
| 141 |
+
MOD_DIR.mkdir(parents=True, exist_ok=True)
|
| 142 |
+
|
| 143 |
+
generated_video = Path(BASE_DIR, "final_video.mp4")
|
| 144 |
+
|
| 145 |
+
# Upload the video
|
| 146 |
+
uploaded_file = st.file_uploader("Upload a small video containing Balloons", type=["mp4"])
|
| 147 |
+
if uploaded_file is not None:
|
| 148 |
+
st.video(uploaded_file)
|
| 149 |
+
vid = uploaded_file.name
|
| 150 |
+
st.info(f'Uploaded {vid}')
|
| 151 |
+
with open(vid, mode='wb') as f:
|
| 152 |
+
f.write(uploaded_file.read())
|
| 153 |
+
uploaded_video = Path(BASE_DIR, vid)
|
| 154 |
+
|
| 155 |
+
# Detect balloon in the frames and generate video
|
| 156 |
+
try:
|
| 157 |
+
video_to_frames(uploaded_video, FRAMES_DIR)
|
| 158 |
+
predict_on_frames(FRAMES_DIR, MOD_DIR)
|
| 159 |
+
frames_to_video(MOD_DIR, generated_video)
|
| 160 |
+
st.success("Successfully Generated!!")
|
| 161 |
+
|
| 162 |
+
# Video file Generated
|
| 163 |
+
video_file = open(str(generated_video), 'rb')
|
| 164 |
+
video_bytes = video_file.read()
|
| 165 |
+
st.video(video_bytes)
|
| 166 |
+
st.download_button('Download the Video', video_bytes, file_name=generated_video.name)
|
| 167 |
+
except Exception as e:
|
| 168 |
+
st.error(f"Could not convert the file due to {e}")
|
| 169 |
+
else:
|
| 170 |
+
st.info('File Not Uploaded Yet!!!')
|
requirements.txt
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
numpy==1.23.3
|
| 2 |
+
pillow==9.2.0
|
| 3 |
+
opencv-python==4.6.0.66
|
| 4 |
+
matplotlib==3.6.1
|
| 5 |
+
torch==1.12.1
|
| 6 |
+
transformers==4.22.2
|
| 7 |
+
stqdm==0.0.4
|