interface / app.py
Jason Adrian
Adding metadata + new class features
4a3b432
raw
history blame
4.55 kB
import gradio as gr
import random
import csv
import datetime
class_names = ['cat', 'dog']
def update_dropdown(className):
class_names.append(className)
updated_choices = gr.Dropdown(choices=class_names)
return updated_choices, updated_choices
def show_picked_class(className):
return className
def image_classifier(inp):
if inp is None:
return {'cat': 0.3, 'dog': 0.7}
num_class = len(class_names)
# Generate random percentages between 0 and 1
percentages = [random.random() for _ in range(num_class)]
total = sum(percentages)
# Normalize the percentages to ensure they sum up to 1
normalized_percentages = [p / total for p in percentages]
labeled_result = {name:score for name, score in zip(class_names, normalized_percentages)}
return labeled_result
demo = gr.Blocks()
with demo as app:
gr.Markdown("# Single Image")
with gr.Row():
with gr.Column():
inp_img = gr.Image()
with gr.Row():
clear_btn = gr.Button(value="Clear")
process_btn = gr.Button(value="Process", variant="primary")
with gr.Column():
out_txt = gr.Label(label="Probabilities", num_top_classes=3)
text_input = gr.Textbox(label="Input the new class here")
b1 = gr.Button("Add new class")
text_options = gr.Dropdown(class_names, label="Class Label", multiselect=False)
b2 = gr.Button("Show me the picked class")
picked_class = gr.Textbox()
b2.click(show_picked_class, inputs=text_options, outputs=picked_class)
process_btn.click(image_classifier, inputs=inp_img, outputs=out_txt)
clear_btn.click(lambda:(
gr.update(value=None),
gr.update(value=None)
),
inputs=None,
outputs=[inp_img, out_txt])
gr.Markdown("# Multiple Images")
def show_to_gallery(images):
file_paths = [[file.name, class_names[0]] for file in images]
# print(file_paths)
return file_paths, file_paths
def get_select_index(evt: gr.SelectData):
# print("data",evt._data)
# print("value",evt.value)
return evt.index
with gr.Column():
imgs = gr.State()
multiple_inputs = gr.UploadButton(label="Upload multiple images file here.", file_count="multiple", file_types=["image"])
gallery = gr.Gallery()
selected = gr.Textbox(label="Image Gallery Index")
images_label = gr.Dropdown(class_names, label="Class Label", multiselect=False)
b3 = gr.Button("Save and change the label using dropdown")
b1.click(update_dropdown, inputs=text_input, outputs=[text_options, images_label])
multiple_inputs.upload(show_to_gallery, inputs=multiple_inputs, outputs=[gallery, imgs])
gallery.select(get_select_index, None, selected)
def change_labels(imgs, index, images_label):
index = int(index)
label_idx = class_names.index(images_label)
imgs[index][1] = class_names[label_idx]
return imgs, imgs
b3.click(change_labels, [imgs, selected, images_label], [imgs, gallery])
gr.Markdown('### Save Metadata Into .csv')
b4 = gr.Button("Upload to metadata")
def upload_metadata(imgs):
time_uploaded = datetime.datetime.now()
time_str = time_uploaded.strftime("%m-%d-%Y_%H-%M-%S")
with open(f'{time_str}.csv', mode='w', newline='') as csv_file:
# Create a CSV writer
csv_writer = csv.writer(csv_file)
# Write the header row
csv_writer.writerow(['image_path', 'ground_truth', 'time_uploaded', 'prediction_label', 'prediction_conf'])
for image in imgs:
image.append(time_str)
model_output = image_classifier(image)
# Sort the label and confidence output in descending order
sorted_output = dict(sorted(model_output.items(), key=lambda item: item[1], reverse=True))
# Extract the label with the highest value
label_prediction = next(iter(sorted_output))
image.append(label_prediction)
label_confidence = model_output[label_prediction]
image.append(label_confidence)
# Write the data rows
csv_writer.writerows(imgs)
print(f"Metadata CSV file has been created.")
b4.click(upload_metadata, inputs=imgs)
demo.launch(debug=True)