Spaces:
Runtime error
Runtime error
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from icevision.all import *
|
2 |
+
import icedata
|
3 |
+
import PIL, requests
|
4 |
+
import torch
|
5 |
+
from torchvision import transforms
|
6 |
+
import gradio as gr
|
7 |
+
|
8 |
+
# Download the dataset
|
9 |
+
url = "https://cvbp-secondary.z19.web.core.windows.net/datasets/object_detection/odFridgeObjects.zip"
|
10 |
+
dest_dir = "fridge"
|
11 |
+
data_dir = icedata.load_data(url, dest_dir)
|
12 |
+
|
13 |
+
# Create the parser
|
14 |
+
|
15 |
+
|
16 |
+
parser = parsers.VOCBBoxParser(annotations_dir="Images/Annotated/augmented", images_dir="Images/Annotated/augmented")
|
17 |
+
|
18 |
+
|
19 |
+
# Parse annotations to create records
|
20 |
+
train_records, valid_records = parser.parse()
|
21 |
+
|
22 |
+
class_map = parser.class_map
|
23 |
+
|
24 |
+
extra_args = {}
|
25 |
+
model_type = models.torchvision.retinanet
|
26 |
+
backbone = model_type.backbones.resnet50_fpn
|
27 |
+
# Instantiate the model
|
28 |
+
model = model_type.model(backbone=backbone(pretrained=True), num_classes=len(parser.class_map), **extra_args)
|
29 |
+
|
30 |
+
# Transforms
|
31 |
+
# size is set to 384 because EfficientDet requires its inputs to be divisible by 128
|
32 |
+
image_size = 640
|
33 |
+
train_tfms = tfms.A.Adapter([*tfms.A.aug_tfms(size=image_size, presize=768), tfms.A.Normalize()])
|
34 |
+
valid_tfms = tfms.A.Adapter([*tfms.A.resize_and_pad(image_size), tfms.A.Normalize()])
|
35 |
+
# Datasets
|
36 |
+
train_ds = Dataset(train_records, train_tfms)
|
37 |
+
valid_ds = Dataset(valid_records, valid_tfms)
|
38 |
+
# Data Loaders
|
39 |
+
train_dl = model_type.train_dl(train_ds, batch_size=8, num_workers=4, shuffle=True)
|
40 |
+
valid_dl = model_type.valid_dl(valid_ds, batch_size=8, num_workers=4, shuffle=False)
|
41 |
+
metrics = [COCOMetric(metric_type=COCOMetricType.bbox)]
|
42 |
+
learn = model_type.fastai.learner(dls=[train_dl, valid_dl], model=model, metrics=metrics)
|
43 |
+
|
44 |
+
learn = learn.load('model')
|
45 |
+
|
46 |
+
import os
|
47 |
+
for root, dirs, files in os.walk(r'sample_images/'):
|
48 |
+
for filename in files:
|
49 |
+
print(filename)
|
50 |
+
|
51 |
+
examples = ["sample_images/"+file for file in files]
|
52 |
+
article="<p style='text-align: center'><a href='https://dicksonneoh.com/fridge-detector/' target='_blank'>Blog post</a></p>"
|
53 |
+
enable_queue=True
|
54 |
+
|
55 |
+
|
56 |
+
#examples = [['sample_images/3.jpg']]
|
57 |
+
examples = [["sample_images/"+file] for file in files]
|
58 |
+
|
59 |
+
def show_preds(input_image, display_label, display_bbox, detection_threshold):
|
60 |
+
|
61 |
+
if detection_threshold==0: detection_threshold=0.5
|
62 |
+
|
63 |
+
img = PIL.Image.fromarray(input_image, 'RGB')
|
64 |
+
|
65 |
+
pred_dict = model_type.end2end_detect(img, valid_tfms, model, class_map=class_map, detection_threshold=detection_threshold,
|
66 |
+
display_label=display_label, display_bbox=display_bbox, return_img=True,
|
67 |
+
font_size=16, label_color="#FF59D6")
|
68 |
+
|
69 |
+
return pred_dict['img']
|
70 |
+
|
71 |
+
# display_chkbox = gr.inputs.CheckboxGroup(["Label", "BBox"], label="Display", default=True)
|
72 |
+
display_chkbox_label = gr.inputs.Checkbox(label="Label", default=True)
|
73 |
+
display_chkbox_box = gr.inputs.Checkbox(label="Box", default=True)
|
74 |
+
|
75 |
+
detection_threshold_slider = gr.inputs.Slider(minimum=0, maximum=1, step=0.1, default=0.5, label="Detection Threshold")
|
76 |
+
|
77 |
+
outputs = gr.outputs.Image(type="pil")
|
78 |
+
|
79 |
+
# Option 1: Get an image from local drive
|
80 |
+
gr_interface = gr.Interface(fn=show_preds, inputs=["image", display_chkbox_label, display_chkbox_box, detection_threshold_slider], outputs=outputs, title='IceApp - Fridge Object', article=article, examples=examples)
|
81 |
+
|
82 |
+
# # Option 2: Grab an image from a webcam
|
83 |
+
# gr_interface = gr.Interface(fn=show_preds, inputs=["webcam", display_chkbox_label, display_chkbox_box, detection_threshold_slider], outputs=outputs, title='IceApp - COCO', live=False)
|
84 |
+
|
85 |
+
# # Option 3: Continuous image stream from the webcam
|
86 |
+
# gr_interface = gr.Interface(fn=show_preds, inputs=["webcam", display_chkbox_label, display_chkbox_box, detection_threshold_slider], outputs=outputs, title='IceApp - COCO', live=True)
|
87 |
+
|
88 |
+
|
89 |
+
gr_interface.launch(inline=False, share=True, debug=True)
|