Add interface

app.py

@@ -0,0 +1,124 @@
+import base64
+import json
+import os
+import re
+import time
+import uuid
+from io import BytesIO
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+import streamlit as st
+from PIL import Image
+from streamlit_drawable_canvas import st_canvas
+from utils import transform_annotations, inference, transform_scale, get_heatmap
+
+
+
+def main():
+    if "button_id" not in st.session_state:
+        st.session_state["button_id"] = ""
+    if "color_to_label" not in st.session_state:
+        st.session_state["color_to_label"] = {}
+
+    color_annotation_app()
+
+    with st.sidebar:
+        st.markdown("---")
+        st.markdown(
+            '<h6>Made in &nbsp<img src="https://streamlit.io/images/brand/streamlit-mark-color.png" alt="Streamlit logo" height="16">&nbsp forked from <a href="https://twitter.com/andfanilo">@andfanilo</a></h6>',
+            unsafe_allow_html=True,
+        )
+        st.markdown(
+            '<div style="margin: 0.75em 0;"><a href="https://www.buymeacoffee.com/andfanilo" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/default-orange.png" alt="Buy Me A Coffee" height="41" width="174"></a></div>',
+            unsafe_allow_html=True,
+        )
+
+def full_app():
+    st.sidebar.header("Configuration")
+    st.markdown(
+        """
+    Draw on the canvas, get the drawings back to Streamlit!
+    * Configure canvas in the sidebar
+    * In transform mode, double-click an object to remove it
+    * In polygon mode, left-click to add a point, right-click to close the polygon, double-click to remove the latest point
+    """
+    )
+
+def color_annotation_app():
+    st.markdown(
+        """
+    #
+    """
+    )
+    
+    try:
+        bg_image = Image.open("img/annotation.jpeg")
+    except:
+        bg_image = None 
+
+    # bg_image = None 
+    image_file = st.file_uploader('Upload your image', type=['jpg', 'jpeg', 'png'], accept_multiple_files=False)
+    if image_file is not None: 
+        bg_image = Image.open(image_file).convert('RGB')
+        bg_image.save('img/annotation.jpeg')
+
+    label_color = (
+        st.sidebar.color_picker("Annotation color: ", "#EA1010") + "77"
+    )  # for alpha from 00 to FF
+    label = st.sidebar.text_input("Label", "Default")
+    mode = "transform" if st.sidebar.checkbox("Adjust ROIs", False) else "rect"
+
+    if bg_image:
+        canvas_result = st_canvas(
+            fill_color=label_color,
+            stroke_width=3,
+            background_image=bg_image,
+            height=320,
+            width=512,
+            drawing_mode=mode,
+            key="color_annotation_app",
+        )
+        if canvas_result.json_data is not None:
+            df = pd.json_normalize(canvas_result.json_data["objects"])
+            df = transform_scale(df, bg_image.size)
+            n_objects = len(canvas_result.json_data['objects'])
+            if  n_objects < 3:
+                st.write('annotate at least {} more object(s)'.format(3 - n_objects))
+                pass
+            else:
+                with st.form("my_form"):
+
+                    # Every form must have a submit button.
+                    count_button_clicked = st.form_submit_button("Count objects")
+                    heatmap_button_clicked = st.form_submit_button("Show heatmaps")
+                    if count_button_clicked:
+                        annotations = transform_annotations(df)
+                        prediction = inference(annotations)
+                        st.write(prediction)
+                    elif heatmap_button_clicked:
+                        annotations = transform_annotations(df)
+                        prediction, heatmap = get_heatmap(annotations)
+                        st.write(f"predicted count: {prediction}")
+                        st.image(heatmap)
+
+
+            if len(df) == 0:
+                return
+            st.session_state["color_to_label"][label_color] = label
+            df["label"] = df["fill"].map(st.session_state["color_to_label"])
+            st.dataframe(df[["top", "left", "width", "height", "fill", "label"]])
+            # st.dataframe(df)
+
+def count_objects():
+    st.write('bla bla')
+
+
+if __name__ == "__main__":
+    st.set_page_config(
+        page_title="Class-agnostic Counting Model", page_icon=":pencil2:"
+    )
+    st.title("Class-agnostic Counting Demo")
+    st.sidebar.subheader("Configuration")
+    main()

packages.txt

@@ -0,0 +1,3 @@
+freeglut3-dev
+libgtk2.0-dev
+libgl1-mesa-glx

requirements.txt

@@ -0,0 +1,6 @@
+streamlit>=0.88
+streamlit-drawable-canvas>=0.8
+svgpathtools
+svgwrite
+opencv-python-headless
+matplotlib

saved_state.json

@@ -0,0 +1,41 @@
+{
+  "version": "4.4.0",
+  "objects": [
+    {
+      "type": "circle",
+      "version": "4.4.0",
+      "originX": "left",
+      "originY": "center",
+      "left": 52,
+      "top": 197,
+      "width": 91.02,
+      "height": 91.02,
+      "fill": "rgba(255, 165, 0, 0.2)",
+      "stroke": "black",
+      "strokeWidth": 5,
+      "strokeDashArray": null,
+      "strokeLineCap": "butt",
+      "strokeDashOffset": 0,
+      "strokeLineJoin": "miter",
+      "strokeUniform": false,
+      "strokeMiterLimit": 4,
+      "scaleX": 1,
+      "scaleY": 1,
+      "angle": 54.39,
+      "flipX": false,
+      "flipY": false,
+      "opacity": 1,
+      "shadow": null,
+      "visible": true,
+      "backgroundColor": "",
+      "fillRule": "nonzero",
+      "paintFirst": "fill",
+      "globalCompositeOperation": "source-over",
+      "skewX": 0,
+      "skewY": 0,
+      "radius": 45.510987684294435,
+      "startAngle": 0,
+      "endAngle": 6.283185307179586
+    }
+  ]
+}

utils.py

@@ -0,0 +1,67 @@
+import requests
+import json
+from PIL import Image
+from io import BytesIO, StringIO
+from base64 import decodebytes
+import numpy as np
+import cv2
+import matplotlib.pyplot as plt
+import os
+
+inference_url = 'ubuntu@ec2-108-137-44-0.ap-southeast-3.compute.amazonaws.com'
+
+def transform_annotations(df):
+    df['x1'] = df['left'] 
+    df['x2'] = (df['left'] + df['width'])
+    df['y1'] = df['top'] 
+    df['y2'] = (df['top'] + df['height']) 
+
+    df['top_left'] = df.apply(lambda x: list([x['x1'], x['y1']]), axis=1) 
+    df['bottom_left'] = df.apply(lambda x: list([x['x1'], x['y2']]), axis=1) 
+    df['bottom_right'] = df.apply(lambda x: list([x['x2'], x['y2']]), axis=1) 
+    df['top_right'] = df.apply(lambda x: list([x['x2'], x['y1']]), axis=1) 
+    annotations = df[['top_left', 'bottom_left', 'bottom_right', 'top_right']].values.tolist()
+    return annotations 
+
+def transform_scale(df, size):
+    if len(df) > 0:
+        x_ratio = 512/ size[0]
+        y_ratio = 320 / size[1]
+
+        df['left'] = df['left'] / x_ratio
+        df['width'] = df['width'] / x_ratio
+        df['top'] = df['top'] / y_ratio
+        df['height'] = df['height'] / y_ratio
+
+    return df
+
+
+def inference(annotations):
+    result = requests.post(
+            f"http://{inference_url}:80/predict",
+        files = {'file': open(f"img/annotation.jpeg", 'rb'),
+                 'data': json.dumps({'annotations': annotations})
+                },
+    )
+
+    return result.json()
+
+def get_heatmap(annotations):
+    result = requests.post(
+            f"http://{inference_url}:80/heatmap",
+        files = {'file': open(f"img/annotation.jpeg", 'rb'),
+                 'data': json.dumps({'annotations': annotations})
+                },
+    )
+
+
+    heatmap = result.json()['heatmap']
+
+    heatmap = np.array(heatmap) * 0.1
+
+
+    colormap = plt.get_cmap('inferno')
+    heatmap = (colormap(heatmap)).astype(np.float32)[:,:,:3]
+
+
+    return result.json()['count'], heatmap