MosquitoAlert Challenge 2023
MosquitoAlert - YoloV5 Baseline Submission
MosquitoAlert - YoloV5 Baseline Submission
dipam
MosquitoAlert - YoloV5 Baseline Submission¶
This notebook will contains instructions and example submission with yolov5 trained on the dataset.
Installations 🤖¶
aicrowd-clifor downloading challenge data and making submissionspyarrowfor saving to parquet for submissions
In [18]:
!pip install -U ultralytics tqdm opencv-python torchvision pandas Pillow gdown aicrowd-cli
In [1]:
import torch
import os
from pathlib import Path
import cv2
from timeit import default_timer as timer
import pandas as pd
from tqdm.auto import tqdm
from PIL import Image
Download Dataset and Models¶
In [3]:
!aicrowd login
In [10]:
!aicrowd dataset download --challenge mosquitoalert-challenge-2023
In [11]:
!unzip -qq test_images_phase1.zip -d test_images_phase1/
In [12]:
# Downloading the weight file at https://drive.google.com/file/d/1s7EoK3V9YTwQOxhuO1ERABRZ95atz8PS/view
!gdown "1s7EoK3V9YTwQOxhuO1ERABRZ95atz8PS"
In [4]:
TEST_DATA_DIR = 'test_images_phase1/'
In [5]:
def classify_image(image):
image_information = {}
result = yolov5_model(image)
result_df = result.pandas().xyxy[0]
if result_df.empty:
print('No results from yolov5 model!')
else:
image_information = result_df.to_dict()
return image_information
# getting mosquito_class name from predicted result
def extract_predicted_mosquito_class_name(extractedInformation):
mosquito_class = ""
if extractedInformation is not None:
mosquito_class = str(extractedInformation.get("name").get(0))
return mosquito_class
# getting mosquito_class number from predicted result
def extract_predicted_mosquito_class_number(extractedInformation):
mosquito_class = ""
if extractedInformation is not None:
mosquito_class = str(extractedInformation.get("class").get(0))
return mosquito_class
# getting mosquito_class confidence score from predicted result
def extract_predicted_mosquito_class_confidence(extractedInformation):
mosquito_class = ""
if extractedInformation is not None:
mosquito_class = str(extractedInformation.get("confidence").get(0))
return mosquito_class
# getting mosquito bounding box from predicted result
# it looks like the results are in different notation
# than the uoutput file
# Pascal VOC? (top, left, bottom, right)?
def extract_predicted_mosquito_bbox(extractedInformation):
bbox = []
if extractedInformation is not None:
xmin = str(extractedInformation.get("xmin").get(0))
ymin = str(extractedInformation.get("ymin").get(0))
xmax = str(extractedInformation.get("xmax").get(0))
ymax = str(extractedInformation.get("ymax").get(0))
bbox = [xmin, ymin, xmax, ymax]
return bbox
# find image id
def find_image_id(original_image):
image_name = os.path.splitext(original_image)[0]
return image_name
In [6]:
# path to dataset
root_images = os.path.join(TEST_DATA_DIR)
all_images = os.listdir(root_images)
print(f"Total images: {len(all_images)}")
class_labels = {
"aegypti": 0,
"albopictus": 1,
"anopheles": 2,
"culex": 3,
"culiseta": 4,
"japonicus/koreicus": 5
}
# counter for correctly recognized mosquito classes
counter = 0
labels_counter = 0
no_mosquito_detected = 0
rows = []
# yolov5 challenge trained baseline
trained_model_pretrained = "mosquitoalert-yolov5-baseline.pt"
# yolov5_model
# Colab gets an error on this -> Use Runtime -> Restart Session (The data will not be deleted)
yolov5_model = torch.hub.load('ultralytics/yolov5', 'custom', path = trained_model_pretrained, force_reload = True)
for original_image in tqdm(all_images):
try:
original_image_file = os.path.join(root_images, original_image)
# checking if it is a file
if os.path.isfile(original_image_file):
# opening testing image
# print(f'You are watching: {original_image}')
# classifying image by yolov5 model
predictedInformation = classify_image(original_image_file)
mosquito_class_name_predicted = ""
mosquito_class_number_predicted = ""
mosquito_class_confidence = 0
mosquito_class_bbox = [0, 0, 0, 0]
if predictedInformation:
mosquito_class_name_predicted = extract_predicted_mosquito_class_name(predictedInformation)
mosquito_class_number_predicted = extract_predicted_mosquito_class_number(predictedInformation)
mosquito_class_confidence = extract_predicted_mosquito_class_confidence(predictedInformation)
mosquito_class_bbox = extract_predicted_mosquito_bbox(predictedInformation)
counter += 1
else:
no_mosquito_detected += 1
# print(f"Predicted mosquito class: {mosquito_class_name_predicted} with {float(mosquito_class_confidence):.2f} confidence score.")
# bbox = [xmin, ymin, xmax, ymax]
# create row for csv file
row = [original_image, mosquito_class_name_predicted, mosquito_class_number_predicted, mosquito_class_confidence,
mosquito_class_bbox[0], mosquito_class_bbox[1], mosquito_class_bbox[2], mosquito_class_bbox[3]]
rows.append(row)
labels_counter += 1
# print(f"Finished file reading, file {original_image} read correctly!")
except Exception as e:
print(f'Unable to process file: {original_image}!')
print(f'Exception: {e}!')
In [7]:
df = pd.DataFrame(rows, columns=["img_fName", "class_label", "predicted_class_number", "confidence_score", "bbx_xtl", "bbx_ytl", "bbx_xbr", "bbx_ybr"])
sub_df = df.drop(['predicted_class_number', 'confidence_score'], axis=1, inplace=False)
sub_df.head()
Out[7]:
In [8]:
sub_df.to_csv('submission_phase1.csv', index=False)
In [9]:
!aicrowd submission create -c mosquitoalert-challenge-2023 -f "submission_phase1.csv"
Content
Comments
You must login before you can post a comment.
Thanks for the trained model! :)
pls , are we required to training the model from scratch or fine tune a pretrained object detection model such as yolo v5
Where is the downloaded data located? when I try to unzip the file, it says : unzip: cannot find or open test_images_phase1.zip, test_images_phase1.zip.zip or test_images_phase1.zip.ZIP.
@adegboyega_samuel you can tried both approaches. You can train model from scratch or fine tune already pretrained model. This notebook is just an example.
@mfalk Could you provide us with the training configuration and whether the entire training dataset was used to train the Yolov5 baseline submission model.
Comment deleted by mfalk.
@saidinesh_pola I will be talking about baseline in tomorrow’s town hall event, if are not able to attend it is going to be available later as a recording. Here is an announcement about townhall: https://discourse.aicrowd.com/t/join-mosquito-alert-challenge-townhall/9080