|
import gradio as gr |
|
import numpy as np |
|
import time |
|
|
|
|
|
def predict(x): |
|
return np.fliplr(x) |
|
|
|
|
|
def compress(): |
|
time.sleep(1) |
|
return 'The model has been compressed successfully.' |
|
|
|
|
|
with gr.Blocks() as demo: |
|
with gr.Column(): |
|
gr.Radio(["Image classification", "Object detection", "Semantic segmentation"], label="Tasks"), |
|
gr.Radio(["ResNet", "VGG", "MobileNet"], label="Models"), |
|
gr.Radio(["Weight quantization","Knowledge distillation","Network pruning", "Neural Architecture Search"], |
|
label="Compression methods"), |
|
gr.Radio(["Jetson Nano"], label="Deployments") |
|
compress_btn = gr.Button("compress") |
|
output_compress = gr.Textbox(lines=1, label="Model Compression Results", visible=False) |
|
|
|
with gr.Row(): |
|
Original_config = gr.Dataframe(headers=["#Params.(M)", "FLOPs(G)"], datatype=[ |
|
"str", "str"], row_count=1, value=[['63.8M','250G']], label="Original model config", visible=False) |
|
Compressed_config = gr.Dataframe(headers=["#Params.(M)", "FLOPs(G)"], datatype=[ |
|
"str", "str"], row_count=1, value=[['34.6M','126G']],label="Compressed model config", visible=False) |
|
with gr.Row(): |
|
input_predict = gr.Image(label="input") |
|
output_predict = gr.Image(label="output") |
|
predict_btn = gr.Button("predict") |
|
state = gr.State() |
|
compress_btn.click(fn=compress, inputs=None, |
|
outputs=output_compress, api_name="compress") |
|
compress_btn.click(lambda : (gr.Textbox.update(visible=True), "visible"), None, [output_compress, state]) |
|
output_compress.change(lambda: (Original_config.update(visible=True), "visible"), None, [Original_config, state]) |
|
output_compress.change(lambda: (Compressed_config.update(visible=True), "visible"), None, [Compressed_config, state]) |
|
predict_btn.click(fn=predict, inputs=input_predict, |
|
outputs=output_predict, api_name="predict") |
|
|
|
demo.launch(share=True) |