def __init__(self):
if torch.cuda.is_available():
device_type = 'cuda:0'
else:
device_type = 'cpu'
if device_type == 'cuda:0':
CPU = False
model_path = '../../models/hardhat.pkl'
self.model = load_learner(model_path) # , cpu=CPU
def predict(self, filename: str, plant: str):
'''classifying image.'''
print("-------------Model Downloading-------------")
self.download_model(plant)
# model = load_learner(Path.cwd()/'../models/export.pkl')
print("-------------Model Downloaded-------------")
model = load_learner(f"{plant}.pkl")
img = PILImage.create(filename)
pred_class, pred_idx, ful_tensor = model.predict(img)
return str(pred_class)
async def setup_learner():
await download_file(model_file_download_url, model_file)
try:
learner = load_learner(model_file)
return learner
except RuntimeError as e:
if len(e.args) > 0 and 'CPU-only machine' in e.args[0]:
print(e)
message = "\n\nThis model was trained with an old version of fastai and will not work in a CPU environment.\n\nPlease update the fastai library in your training environment and export your model again.\n\nSee instructions for 'Returning to work' at https://course.fast.ai."
raise RuntimeError(message)
else:
raise
from fastai.vision.all import load_learner, torch, PILImage
import streamlit as st
import os
import time
st.set_page_config("PETS", "🐶&🐱")
st.title("PETS 🐶&🐱")
learn = load_learner('models/pets.pkl')
def predict(img):
st.image(img, use_column_width=True)
with st.spinner('Wait for it...'):
time.sleep(3)
clas, clas_idx, probs = learn.predict(img)
prob = round(torch.max(probs).item() * 100, 2)
st.success(f"This is {clas} with proability of {prob}%.")
option = st.radio('', ['Choose a test image', 'Choose your own image'])
if option == 'Choose a test image':
test_images = os.listdir('images/')
test_image = st.selectbox('Please select a test image:', test_images)
file_path = 'images/' + test_image
img = PILImage.create(file_path)
predict(img)
else:
import json
import time
from io import BytesIO
import requests
import numpy as np
from fastai.vision.all import load_learner, PILImage
learn = load_learner('export.pkl')
def lambda_handler(event, context):
"""Sample pure Lambda function
Parameters
----------
event: dict, required
API Gateway Lambda Proxy Input Format
Event doc: https://docs.aws.amazon.com/apigateway/latest/developerguide/set-up-lambda-proxy-integrations.html#api-gateway-simple-proxy-for-lambda-input-format
context: object, required
Lambda Context runtime methods and attributes
Context doc: https://docs.aws.amazon.com/lambda/latest/dg/python-context-object.html
Returns
------
API Gateway Lambda Proxy Output Format: dict
Return doc: https://docs.aws.amazon.com/apigateway/latest/developerguide/set-up-lambda-proxy-integrations.html
"""
def __init__(self):
self.model = load_learner('webapp/HWES/utils/export.pkl')
self.math = Math()
self.the_list = []
)
import glob
import streamlit as st
from PIL import Image
from random import shuffle
import urllib.request
# set images
image1 = Image.open('sidebar1.jpg')
image2 = Image.open('menu1.jpg')
image3 = Image.open('menu2.jpg')
#MODEL_URL = "https://github.com/GemmyTheGeek/FoodyDudy/raw/main/resnet34-10.pkl"
MODEL_URL = "http://freelyblog.com/resnet34-10.pkl"
urllib.request.urlretrieve(MODEL_URL, "model.pkl")
learn_inf = load_learner('model.pkl', cpu=True)
thaimenu = [
"แกงเขียวหวานไก่", "แกงเทโพ", "แกงเลียง", "แกงจืดเต้าหู้หมูสับ",
"แกงจืดมะระยัดไส้", "แกงมัสมั่นไก่", "แกงส้มกุ้ง",
"ไก่ผัดเม็ดมะม่วงหิมพานต์", "ไข่เจียว", "ไข่ดาว", "ไข่พะโล้", "ไข่ลูกเขย",
"กล้วยบวชชี", "ก๋วยเตี๋ยวคั่วไก่", "กะหล่ำปลีผัดน้ำปลา", "กุ้งแม่น้ำเผา",
"กุ้งอบวุ้นเส้น", "ขนมครก", "ข้าวเหนียวมะม่วง", "ข้าวขาหมู",
"ข้าวคลุกกะปิ", "ข้าวซอยไก่", "ข้าวผัด", "ข้าวผัดกุ้ง", "ข้าวมันไก่",
"ข้าวหมกไก่", "ต้มข่าไก่", "ต้มยำกุ้ง", "ทอดมัน", "ปอเปี๊ยะทอด",
"ผักบุ้งไฟแดง", "ผัดไท", "ผัดกะเพรา", "ผัดซีอิ๊วเส้นใหญ่",
"ผัดฟักทองใส่ไข่", "ผัดมะเขือยาวหมูสับ", "ผัดหอยลาย", "ฝอยทอง", "พะแนงไก่",
"ยำถั่วพู", "ยำวุ้นเส้น", "ลาบหมู", "สังขยาฟักทอง", "สาคูไส้หมู", "ส้มตำ",
"หมูปิ้ง", "หมูสะเต๊ะ", "ห่อหมก"
]
def playground():
st.title("Playground")
# Initial instructions
init_info = st.empty()
init_info.info(
"ℹ️ Upload an image on the sidebar to run the model on it!\n"
"Just be sure that it's satellite imagery, otherwise you're "
"just going to get random outputs 🤷♂️")
# Set the sidebar inputs
st.sidebar.title("Inputs")
# NOTE For the sake of time, we're just going to use the `Land scenes` model
# model_type = st.sidebar.radio("Model", ["Deforestation", "Land scenes"])
model_type = "Land scenes"
input_file = st.sidebar.file_uploader(
"Upload an image",
type=["jpg", "jpeg", "png", "tif"],
accept_multiple_files=False,
help="Test our model on an image of your liking! "
"But remember that this should only work for satellite imagery, "
"ideally around 256x256 size.",
)
st.sidebar.markdown(
"Made by [André Ferreira](https://andrecnf.com/) and [Karthik Bhaskar](https://www.kbhaskar.com/)."
)
# Set the model
model = load_learner(
"fsdl_deforestation_detection/modeling/resnet50-128.pkl")
# Speed up model inference by deactivating gradients
model.model.eval()
torch.no_grad()
if input_file is not None:
# Load and display the uploaded image
with st.spinner("Loading image..."):
img = imread(input_file)
# Check if it's a different image than the one before
if input_file.name != session_state.image_name:
session_state.image_name = input_file.name
session_state.image_id = gen_image_id()
session_state.ts = datetime.now()
session_state.user_data_uploaded = False
# Reset buttons
session_state.user_feedback_positive = False
session_state.user_feedback_negative = False
fig = px.imshow(img)
st.plotly_chart(fig)
init_info.empty()
# Run the model on the image
output = run_model(model, img)
st.subheader("Model output:")
show_model_output(model_type, output)
st.info(
"ℹ️ Green labels represent categories that we don't associate with deforestation "
"risk (e.g. natural occurences or old structures), while red labels can serve as "
"a potential deforestation signal (e.g. new constructions, empty patches in forests)."
)
# User feedback / data flywheel
st.write("Did the model output match what you expected?")
feedback_cols = st.beta_columns(2)
with feedback_cols[0]:
positive_btn = st.button("✅")
with feedback_cols[1]:
negative_btn = st.button("❌")
if (positive_btn
or session_state.user_feedback_positive) and not negative_btn:
session_state.user_feedback_positive = True
session_state.user_feedback_negative = False
st.info("ℹ️ Thank you for your feedback! This can help us "
"improve our models 🙌")
if session_state.user_data_uploaded is False:
upload_user_data(
session_state.user_id,
session_state.ts,
session_state.image_id,
img,
output[1],
session_state.user_feedback_positive,
)
session_state.user_data_uploaded = True
if st.button("Delete my image and feedback data"):
st.info("ℹ️ Alright, we deleted it. Just know that we had "
"high expectations that you could help us improve "
"deforestation detection models. We thought we "
"were friends 🙁")
elif (negative_btn
or session_state.user_feedback_negative) and not positive_btn:
session_state.user_feedback_positive = False
session_state.user_feedback_negative = True
st.info("ℹ️ Thank you for your feedback! This can help us "
"improve our models 🙌\n"
"It would be even better if you could tell us "
"what makes you think the model failed. Mind "
"leaving a comment bellow?")
if session_state.user_data_uploaded is False:
upload_user_data(
session_state.user_id,
session_state.ts,
session_state.image_id,
img,
output[1],
session_state.user_feedback_positive,
)
session_state.user_data_uploaded = True
user_comment = st.empty()
user_comment_txt = user_comment.text_input(
label="Leave a comment on why the model failed.",
max_chars=280,
)
if len(user_comment_txt) > 0:
upload_user_comment(
session_state.user_id,
session_state.image_id,
user_comment_txt,
)
if st.button("Delete my image and feedback data"):
st.info("ℹ️ Alright, we deleted it. Just know that we had "
"high expectations that you could help us improve "
"deforestation detection models. We thought we "
"were friends 🙁")
delete_user_data(session_state.user_id, session_state.image_id)
# Model interpretation
with st.beta_expander("Peek inside the black box"):
explain_cols = st.beta_columns(2)
with explain_cols[0]:
st.subheader("Model structure")
st.info(
"ℹ️ Our model is largely based on the [ResNet](https://paperswithcode.com/method/resnet) "
"archirtecture, using a ResNet50 from [FastAI](https://docs.fast.ai/). "
"Bellow you can see the model's layer definition.")
st.text(model.model)
with explain_cols[1]:
st.subheader("Output interpretation")
# TODO Add the result of applying SHAP to the model in the current sample
st.info(
"ℹ️ Given some difficulties with using [SHAP](https://github.com/slundberg/shap) "
"with [FastAI](https://docs.fast.ai/), we haven't implemented this yet. "
"Would you like to give it a try?")
def overview():
st.title("Overview")
# Initial instructions
init_info = st.empty()
# Set the sidebar inputs
st.sidebar.title("Inputs")
# NOTE For the sake of time, we're just going to use the `Land scenes` model
# model_type = st.sidebar.radio("Model", ["Deforestation", "Land scenes"])
model_type = "Land scenes"
dataset_name = st.sidebar.radio("Dataset", ["Amazon", "Oil palm"])
chosen_set = None
if dataset_name == "Amazon":
chosen_set = st.sidebar.radio("Set", ["Train", "Validation"])
init_info.info(
"ℹ️ You've selected the "
"[Amazon dataset](https://www.kaggle.com/c/planet-understanding-the-amazon-from-space/), "
"which is the one in which our models were trained on. As such, you can look at performance "
"on either the train or validation set.")
(
bucket_name,
img_path,
labels_table,
img_name_col,
label_col,
) = set_paths(dataset_name, model_type)
else:
init_info.info(
"ℹ️ You've selected the "
"[oil palm dataset](https://www.kaggle.com/c/widsdatathon2019/), "
"which can be seen as a test dataset, i.e. it wasn't used during training. "
"While it should be somewhat similar to the Amazon dataset, it can be interesting "
"to compare results on potentially out-of-domain data.")
(
bucket_name,
img_path,
labels_table,
img_name_col,
label_col,
) = set_paths(dataset_name, model_type)
# Set the model
model = load_learner(
"fsdl_deforestation_detection/modeling/resnet50-128.pkl")
# Speed up model inference by deactivating gradients
model.model.eval()
torch.no_grad()
img_names = load_image_names(model, chosen_set, bucket_name, labels_table)
sample_name = st.sidebar.selectbox(
"Sample",
img_names,
)
st.sidebar.markdown(
"Made by [André Ferreira](https://andrecnf.com/) and [Karthik Bhaskar](https://www.kbhaskar.com/)."
)
# Load all the data (or some samples) from the selected database
n_samples = 250
imgs, labels = load_data(
dataset_name,
model_type,
bucket_name,
img_path,
img_names,
labels_table,
img_name_col,
n_samples=n_samples,
)
# Show some performance metrics
# TODO Use all the set data to get the correct performance metrics
st.header("Performance")
metrics_cols = st.beta_columns(2)
with st.spinner("Getting performance results..."):
if dataset_name == "Amazon":
# NOTE This are the metrics obtained for the validation set,
# when training the model in Colab; ideally, this should still
# be calculated dynamically in here, but it's proving to be
# slow and impractical in the approach that we were taking
acc = 0.956407
fbeta = 0.926633
# pred, acc, fbeta = get_performance_metrics(
# model, imgs, labels, dataset_name
# )
pred, _, _ = get_performance_metrics(model, imgs, labels,
dataset_name)
else:
pred, acc, fbeta = get_performance_metrics(model, imgs, labels,
dataset_name)
acc, fbeta = 100 * acc, 100 * fbeta
with metrics_cols[0]:
fig = get_gauge_plot(acc, title="Accuracy")
st.plotly_chart(fig, use_container_width=True)
with metrics_cols[1]:
fig = get_gauge_plot(fbeta, title="F2")
st.plotly_chart(fig, use_container_width=True)
if dataset_name == "Amazon":
st.info(
"ℹ️ These are the validation metrics [obtained when training the model](https://colab.research.google.com/github/karthikraja95/fsdl_deforestation_detection/blob/master/fsdl_deforestation_detection/experimental/FSDL_Final_Model.ipynb)."
)
else:
st.info(
"ℹ️ Showing performance metrics here by mapping the original labels to a "
"binary, deforestation label. This should be somewhat relatable to the "
"presence of oil palm plantations, which is the label in this dataset."
)
# Show number of samples
fig = get_number_plot(len(img_names), title="Samples")
st.plotly_chart(fig, use_container_width=True)
# Show label analysis
st.header("Label analysis")
labels_cols = st.beta_columns(2)
with labels_cols[0]:
fig = get_hist_plot(
labels,
"labels",
dataset_name,
model_type,
title="Labels distribution",
)
st.plotly_chart(fig, use_container_width=True)
with labels_cols[1]:
fig = get_hist_plot(
pred,
"predictions",
dataset_name,
model_type,
title="Predicted labels distribution",
)
st.plotly_chart(fig, use_container_width=True)
st.info(
f"ℹ️ Using only a subset of {n_samples} samples, so as to make this plot practically fast."
)
# Show imagery analysis
st.header("Imagery analysis")
st.subheader("Image size")
img_size_cols = st.beta_columns(3)
with img_size_cols[0]:
fig = get_number_plot(imgs.shape[1], title="Height")
st.plotly_chart(fig, use_container_width=True)
with img_size_cols[1]:
fig = get_number_plot(imgs.shape[2], title="Width")
st.plotly_chart(fig, use_container_width=True)
with img_size_cols[2]:
fig = get_number_plot(imgs.shape[3], title="Channels")
st.plotly_chart(fig, use_container_width=True)
fig = get_pixel_dist_plot(imgs)
st.plotly_chart(fig, use_container_width=True)
st.info(
f"ℹ️ Using only a subset of {n_samples} samples, so as to make this plot practically fast."
)
# Show sample analysis
st.header("Sample analysis")
# Load and display the uploaded image
with st.spinner("Loading image..."):
img = load_image(bucket_name, img_path, sample_name)
fig = px.imshow(img)
st.plotly_chart(fig)
# Run the model on the image
output = run_model(model, img)
st.subheader("Model output:")
show_model_output(model_type, output)
st.subheader("Real labels:")
show_labels(dataset_name, sample_name, labels_table, img_name_col,
label_col)
st.info(
"ℹ️ Green labels represent categories that we don't associate with deforestation "
"risk (e.g. natural occurences or old structures), while red labels can serve as "
"a potential deforestation signal (e.g. new constructions, empty patches in forests)."
)
# Model interpretation
with st.beta_expander("Peek inside the black box"):
explain_cols = st.beta_columns(2)
with explain_cols[0]:
st.subheader("Model structure")
st.info(
"ℹ️ Our model is largely based on the [ResNet](https://paperswithcode.com/method/resnet) "
"archirtecture, using a ResNet50 from [FastAI](https://docs.fast.ai/). "
"Bellow you can see the model's layer definition.")
st.text(model.model)
with explain_cols[1]:
st.subheader("Output interpretation")
# TODO Add the result of applying SHAP to the model in the current sample
st.info(
"ℹ️ Given some difficulties with using [SHAP](https://github.com/slundberg/shap) "
"with [FastAI](https://docs.fast.ai/), we haven't implemented this yet. "
"Would you like to give it a try?")
import cv2
from fastai.vision.all import load_learner
import numpy as np
import dlib
import time
start = time.time()
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(
"/home/hang/PycharmProjects/MaskDetector/venv/models"
"/shape_predictor_68_face_landmarks.dat")
learn_inf = load_learner(
'/home/hang/PycharmProjects/MaskDetector/venv/models/risk_v2.pkl')
cap = cv2.VideoCapture(
"/home/hang/PycharmProjects/MaskDetector/venv/Resources/Hang_video_Nov_1.mp4"
)
# font
font = cv2.FONT_HERSHEY_SIMPLEX
# fontScale
fontScale = 0.5
# Line thickness of 2 px
thickness = 1
risk_color = {"low_risk": (0, 255, 0), "high_risk": (0, 0, 255)}
risk_counter = {"low_risk": 0, "high_risk": 0}
prev_time = -1
processed_frames = 0
sample_rate = 1
def load_model():
global model
model = load_learner('CleanMessyModel.pkl')
print('Model loaded')
def predict(filename: str):
'''classifying image.'''
model = load_learner(Path.cwd() / 'bear.pkl')
img = PILImage.create(filename)
pred_class, pred_idx, ful_tensor = model.predict(img)
return str(pred_class)
import streamlit as st
import numpy as np
from PIL import Image
from fastai.vision.all import load_learner, Path
st.title("WELCOME CAT DOG CLASSIFIER")
st.title("cat dog classifer")
uploaded_file = st.file_uploader("Choose an image...", type="jpg")
#file upload
learn_inf = load_learner("exportt.pkl") #load trained model
#classification
if uploaded_file is not None:
#image transformation and prediciton
img = Image.open(uploaded_file)
st.image(img, caption='Your Image.', use_column_width=True)
image = np.asarray(img)
label = learn_inf.predict(image)
#label[0] accesses the actual label string
#output display
st.write("")
st.write("Classifying...")
#check for vowels in the names for correct grammar
if label[0][0] in "AEIOU":
st.write("## This looks like an")
else:
st.write("## This looks like a")
#our labels are in capital letters only
st.title(label[0].lower().title())
from flask import Flask, render_template, request, url_for
from fastai.vision.all import load_learner, PILImage
import re
# from flask_cors import CORS, cross_origin
import logging
app = Flask(__name__)
learn = load_learner('./models/model.pkl')
@app.route('/')
def homepage():
return render_template('index.html')
def predict_single(img_file):
'''function to take image and return prediction'''
pred, pred_idx, probs = learn.predict(PILImage.create(img_file))
probs_list = [pred, probs[pred_idx].tolist()]
return probs_list
@app.route('/result', methods=['POST'])
def predict():
if request.method == 'POST':
my_prediction = predict_single(request.files['inputImage'])
plant_class = str(my_prediction[0])
plant_type = re.search(r'.+?(?=_)', plant_class).group()
growth_stage = re.search(r'(?<=_)[\w+.-]+', plant_class).group()
probability = str("{:.2f}".format(float(my_prediction[1]) * 100))
from envparse import env
from starlette.applications import Starlette
from starlette.staticfiles import StaticFiles
from starlette.responses import JSONResponse
from starlette.templating import Jinja2Templates
import uvicorn
from fastai.vision.all import load_learner
templates = Jinja2Templates(directory='templates')
app = Starlette(debug=True)
app.mount('/static', StaticFiles(directory='statics'), name='static')
model = load_learner('pokemon_classifier.pkl')
@app.route('/')
async def homepage(request):
template = "index.html"
context = {"request": request}
return templates.TemplateResponse(template, context)
@app.route('/classify', methods=['POST'])
async def classify(request):
data = await request.form()
bytes = await (data["file"].read())
prediction, position, tensor = model.predict(bytes)
return JSONResponse({
"prediction": prediction,