traffic_object_detection.py

# -*- coding: utf-8 -*-
"""traffic_object_detection.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1B7DIM9ABIA6RRhA8tL_3rcxL9M1iIP7D
"""

!pip install datasets

from datasets import load_dataset

dataset = load_dataset("Sayali9141/traffic_signal_images")

next(iter(dataset['train']))

import matplotlib.pyplot as plt
from IPython.display import display
from PIL import Image

"""Trying out hugging face YOLO

"""

from transformers import AutoFeatureExtractor

feature_extractor = AutoFeatureExtractor.from_pretrained("hustvl/yolos-small")

from transformers import YolosForObjectDetection

model = YolosForObjectDetection.from_pretrained("hustvl/yolos-small")

"""This code shows how to get image from the url"""

device = 'cuda'

model = model.to(device)

from PIL import Image
import requests
import base64
from io import BytesIO
from time import time
import matplotlib.pyplot as plt
import torch
# colors for visualization
COLORS = [[0.000, 0.447, 0.741], [0.850, 0.325, 0.098], [0.929, 0.694, 0.125],
          [0.494, 0.184, 0.556], [0.466, 0.674, 0.188], [0.301, 0.745, 0.933]]

def plot_results(pil_img, prob, boxes):
    count=0
    plt.figure(figsize=(16,10))
    plt.imshow(pil_img)
    ax = plt.gca()
    colors = COLORS * 100
    for p, (xmin, ymin, xmax, ymax), c in zip(prob, boxes.tolist(), colors):
      cl = p.argmax()
      if model.config.id2label[cl.item()] in ['car', 'truck'] :
          ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin,
                                    fill=False, color=c, linewidth=3))
          text = f'{model.config.id2label[cl.item()]}: {p[cl]:0.2f}'
          ax.text(xmin, ymin, text, fontsize=15,
                  bbox=dict(facecolor='yellow', alpha=0.5))
          count+=1
    plt.axis('off')
    plt.show()
    # print(count)
    return(count)

all_counts = []
for i in range (22000, 22005):
  row = dataset['train'][i]
  start= time()
  pixel_values = feature_extractor(row['image_url'], return_tensors="pt").pixel_values
  pixel_values = pixel_values.to(device)
  # pixel_values.shape
  with torch.no_grad():
    outputs = model(pixel_values, output_attentions=True)
  probas = outputs.logits.softmax(-1)[0, :, :-1]
  keep = probas.max(-1).values > 0.8
  target_sizes = torch.tensor(row['image_url'].size[::-1]).unsqueeze(0)
  postprocessed_outputs = feature_extractor.post_process(outputs, target_sizes)
  bboxes_scaled = postprocessed_outputs[0]['boxes']
  plot_results(row['image_url'], probas[keep], bboxes_scaled[keep])
  count = 0

  for p, boxes in zip(probas[keep], bboxes_scaled[keep]):
    cl = p.argmax()
    if model.config.id2label[cl.item()] in ['car', 'truck']:
      count += 1
  all_counts.append(count)
  print(time()-start)

# def select_columns(example):
#     return {key: example[key] for key in ['timestamp', 'camera_id', 'latitude', 'longitude']}

# subset_dataset = dataset['train'].map(select_columns[dataset['train']])

# data_yolo= subset_dataset.to_pandas()

# data_yolo['box_count'][22000:22004]= [x for x in all_counts]

#create interactive map
#create interactive map using latitude and longitude of counts column
# import folium
# from folium import plugins

# # Create a map object and center it to the avarage coordinates to m
# m = folium.Map(location=[df['latitude'].mean(), df['longitude'].mean()], zoom_start=10)

# # Add marker for each row in the data
# for i in range(0,len(df)):
#     folium.Marker([df.iloc[i]['latitude'], df.iloc[i]['longitude']], popup=df.iloc[i]['counts']).add_to(m)

# # Display the map
# m.save('map.html')
# m