app.py

# Copied megadetector section from https://huggingface.co/spaces/hlydecker/MegaDetector_v5 

# %%
#all imports
import gradio as gr
import torch
import torchvision
import numpy as np
from PIL import Image

# %%
# Loads a model from github repo, but you need to have the model 
model = torch.hub.load('ultralytics/yolov5', 'custom', "https://huggingface.co/spaces/vchiang001/MegaDetector_DLC_pose/blob/main/md_v5b.0.0.pt", force_reload=True)

# %% 

#not sure if we need to resize... maybe just the origin image and see? 
""" def yolo(im): #size=640):
g = (size / max(im.size))  # gain
    im = im.resize((int(x * g) for x in im.size), Image.ANTIALIAS)  # resize 
    
    model = torch.hub.load('ultralytics/yolov5', 'custom', "/home/vic/vic_data/dlclive4mega/models/md_v5b.0.0.pt", force_reload=True)
    
    results = model(im)  # inference
    results.render()  # updates results.imgs with boxes and labels
    return Image.fromarray(results.imgs[0])  """

def yolo(im): #size=640):
    model = torch.hub.load('ultralytics/yolov5', 'custom', "/home/vic/vic_data/dlclive4mega/models/md_v5b.0.0.pt", force_reload=True)
    results = model(im)  # inference
    results.render()  # updates results.imgs with boxes and labels
    return Image.fromarray(results.imgs[0])


# %% 
#inputs = [image, chosen_model, size]
#this is where you show the image as interface 
inputs = gr.inputs.Image(type="pil", label="Input Image")
outputs = gr.outputs.Image(type="pil", label="Output Image")


# %% 
title = "MegaDetector with DeepLabCut pose estimation"
description = "Detect and identify animals, people and vehicles in camera trap images followed by generating poses for humans and animals"
article = "<p style='text-align: center'>Detect and identify animals, people and vehicles in camera trap images followed by generating poses for humans and animals</a></p>"

# %% 
#running the actual 
gr.Interface(yolo, inputs, outputs, title=title, description=description, article=article, theme="huggingface").launch(enable_queue=True)