Trampas_Barcelo / app.py
fcernafukuzaki's picture
Update app.py
b3343ab verified
# -*- coding: utf-8 -*-
"""Deploy Barcelo demo.ipynb
Automatically generated by Colaboratory.
Original file is located at
    https://colab.research.google.com/drive/1FxaL8DcYgvjPrWfWruSA5hvk3J81zLY9
![   ](https://www.vicentelopez.gov.ar/assets/images/logo-mvl.png)
# Modelo
YOLO es una familia de modelos de detección de objetos a escala compuesta entrenados en COCO dataset, e incluye una funcionalidad simple para Test Time Augmentation (TTA), model ensembling, hyperparameter evolution, and export to ONNX, CoreML and TFLite.
## Gradio Inferencia
![](https://i.ibb.co/982NS6m/header.png)
Este Notebook se acelera opcionalmente con un entorno de ejecución de GPU
----------------------------------------------------------------------
 YOLOv5 Gradio demo
*Author: Ultralytics LLC and Gradio*
# Código
"""
#!pip install -qr https://raw.githubusercontent.com/ultralytics/yolov5/master/requirements.txt gradio # install dependencies
import os
import re
import json
import numpy as np
import pandas as pd
import gradio as gr
import torch
from PIL import Image
# Images
torch.hub.download_url_to_file('https://huggingface.co/spaces/Municipalidad-de-Vicente-Lopez/Trampas_Barcelo/resolve/main/2024-03-11T10-50-27.jpg', 'ejemplo1.jpg')
torch.hub.download_url_to_file('https://i.pinimg.com/originals/c2/ce/e0/c2cee05624d5477ffcf2d34ca77b47d1.jpg', 'ejemplo2.jpg')
# model = torch.hub.load('ultralytics/yolov9', 'custom', path='best.pt', force_reload=True, autoshape=True, trust_repo=True)
model = torch.hub.load('/content/yolov9', 'custom', path='/content/yolov9/best.pt', source='local', force_reload=True, autoshape=True)
#model = torch.hub.load('yolov9', 'custom', path='best.pt', source='local', force_reload=True, autoshape=True)  # load on CPU
# Model
class YOLODetect():
def __init__(self, modelo):
self.modelo = modelo
def predecir(self, img, imgsz=640, conf=0.5, iou=0.40):
# iou float 0.7 umbral de intersección sobre unión (IoU) para NMS
# conf float 0.25 umbral de confianza del objeto para la detección
self.modelo.iou = iou
self.modelo.conf = conf
self.results = self.modelo(img) # inference
return self.results
def to_json(self):
detail = []
for index, row in self.results_df.iterrows():
item = {
"quantity": row['Cantidad'],
"description": row['Especie']
}
detail.append(item)
data = {
"image": self.results.files[0],
"size": f"{self.results.s[2]}x{self.results.s[3]}",
"detail": detail
}
return data
def to_dataframe(self):
labels_map = {
'Aedes': "Aedes",
'Mosquito': "Mosquitos",
'Mosca': "Moscas",
}
labels = list(labels_map.keys())
columns_name = {'class': 'Cantidad', 'name': 'Especie'}
self.results_df = self.results.pandas().xyxy[0][['class','name']].groupby('name').count().reset_index().rename(columns=columns_name)
self.results_df = pd.merge(pd.DataFrame(labels, columns=['Especie']), self.results_df, how='left', on='Especie').fillna(0)
self.results_df['Cantidad'] = self.results_df['Cantidad'].astype(int)
self.results_df['Especie'] = self.results_df['Especie'].map(labels_map)
return self.results_df
modelo_yolo = YOLODetect(model)
def yolo(size, iou, conf, im):
'''Wrapper fn for gradio'''
g = (int(size) / max(im.size)) # gain
im = im.resize((int(x * g) for x in im.size), Image.LANCZOS) # resize with antialiasing
im = np.asarray(im, dtype=np.float32)
resultado = modelo_yolo.predecir(im, imgsz=size, conf=conf, iou=iou)
resultado.render() # updates results.imgs with boxes and labels
resultado_df = modelo_yolo.to_dataframe()
resultado_json = modelo_yolo.to_json()
return Image.fromarray(resultado.ims[0]), resultado_df, resultado_json
#------------ Interface-------------
in1 = gr.inputs.Radio(['640', '1280'], label="Tamaño de la imagen", default='640', type='value')
in2 = gr.inputs.Slider(minimum=0, maximum=1, step=0.05, default=0.25, label='NMS IoU threshold')
in3 = gr.inputs.Slider(minimum=0, maximum=1, step=0.05, default=0.50, label='Umbral o threshold')
in4 = gr.inputs.Image(type='pil', label="Original Image")
out2 = gr.outputs.Image(type="pil", label="YOLOv9")
out3 = gr.outputs.Dataframe(label="Cantidad_especie", headers=['Cantidad','Especie'], type="pandas")
out4 = gr.outputs.JSON(label="JSON")
#-------------- Text-----
title = 'Trampas Barceló'
description = """
<p>
<center>
Sistemas de Desarrollado por Subsecretaría de Modernización del Municipio de Vicente López. Advertencia solo usar fotos provenientes de las trampas Barceló, no de celular o foto de internet.
<img src="https://www.vicentelopez.gov.ar/assets/images/logo-mvl.png" alt="logo" width="250"/>
</center>
</p>
"""
article ="<p style='text-align: center'><a href='https://docs.google.com/presentation/d/1T5CdcLSzgRe8cQpoi_sPB4U170551NGOrZNykcJD0xU/edit?usp=sharing' target='_blank'>Para mas info, clik para ir al white paper</a></p><p style='text-align: center'><a href='https://drive.google.com/drive/folders/1owACN3HGIMo4zm2GQ_jf-OhGNeBVRS7l?usp=sharing ' target='_blank'>Google Colab Demo</a></p><p style='text-align: center'><a href='https://github.com/Municipalidad-de-Vicente-Lopez/Trampa_Barcelo' target='_blank'>Repo Github</a></p></center></p>"
examples = [['640',0.25, 0.5,'ejemplo1.jpg'], ['640',0.25, 0.5,'ejemplo2.jpg']]
iface = gr.Interface(yolo,
inputs=[in1, in2, in3, in4],
outputs=[out2,out3,out4], title=title,
description=description,
article=article,
examples=examples,
analytics_enabled=False,
allow_flagging="manual",
flagging_options=["Correcto", "Incorrecto", "Casi correcto", "Error", "Otro"],
#flagging_callback=hf_writer
)
#iface.queue()
iface.launch(server_name="0.0.0.0", server_port=7860, enable_queue=True, debug=True)
#iface.launch(server_name="0.0.0.0", server_port=7860, debug=True)
"""For YOLOv5 PyTorch Hub inference with **PIL**, **OpenCV**, **Numpy** or **PyTorch** inputs please see the full [YOLOv5 PyTorch Hub Tutorial](https://github.com/ultralytics/yolov5/issues/36).
## Citation
[![DOI](https://zenodo.org/badge/264818686.svg)](https://zenodo.org/badge/latestdoi/264818686)
"""