Tree crown detection using DeepForest

import glob
import os
import urllib
import numpy as np

import intake
import matplotlib.pyplot as plt
import xmltodict
import cv2

import torch

from shapely.geometry import box
import pandas as pd
from geopandas import GeoDataFrame
import xarray as xr
import panel as pn
import holoviews as hv
import hvplot.pandas
import hvplot.xarray
from skimage.exposure import equalize_hist

import pooch

import warnings
warnings.filterwarnings(action='ignore')

hv.extension('bokeh', width=100)
%matplotlib inline

for row in target_bbox:
    cv2.rectangle(image2, (int(row["xmin"]), int(row["ymin"])), (int(row["xmax"]), int(row["ymax"])), (0,255,255), thickness=2, lineType=cv2.LINE_AA)

plot_reference = plt.figure(figsize=(15,15))
cv2_imshow(np.flip(image2,2))
plt.title('Reference labels',fontsize='xx-large')
plt.show()

image3 = image.values.copy() 

for index, row in pred_boxes.iterrows():
    cv2.rectangle(image3, (int(row["xmin"]), int(row["ymin"])), (int(row["xmax"]), int(row["ymax"])), (0,255,255), thickness=2, lineType=cv2.LINE_AA)

plot_fullimage = plt.figure(figsize=(15,15))
cv2_imshow(np.flip(image3,2))
plt.title('Full-image predictions',fontsize='xx-large')
plt.show()

plot_referandfullimage = plt.figure(figsize=(15,15))
ax1 = plt.subplot(1, 2, 1), cv2_imshow(np.flip(image2,2))
ax1[0].set_title('Reference labels',fontsize='xx-large')
ax2 = plt.subplot(1, 2, 2), cv2_imshow(np.flip(image3,2))
ax2[0].set_title('Full-image predictions', fontsize='xx-large')
plt.show() # To show figure

#Loop through a few sample windows, crop and predict
plot_tilewindows, axes, = plt.subplots(nrows=2,ncols=2, figsize=(15,15))
axes = axes.flatten()
for index2 in range(4):
    crop = image.values[windows[index2].indices()]
    #predict in bgr channel order, color predictions in red.
    boxes = model.predict_image(image=np.flip(crop[...,::-1],2), return_plot = True)

    #but plot in rgb channel order
    axes[index2].imshow(boxes[...,::-1])
    axes[index2].set_title(f'Prediction in Window {index2 + 1} out of {len(windows)}', fontsize='xx-large')

tile = model.predict_tile(image=image.values,return_plot=False,patch_overlap=0,iou_threshold=0.05,patch_size=400)

# plot predicted bbox
image_tile = image.values.copy()

for index, row in tile.iterrows():
    cv2.rectangle(image_tile, (int(row["xmin"]), int(row["ymin"])), (int(row["xmax"]), int(row["ymax"])), (0, 255, 255), thickness=2, lineType=cv2.LINE_AA)

plot_tilewise = plt.figure(figsize=(15,15))
ax1 = plt.subplot(1, 2, 1), cv2_imshow(np.flip(image2,2))
ax1[0].set_title('Reference labels',fontsize='xx-large')
ax2 = plt.subplot(1, 2, 2), cv2_imshow(np.flip(image_tile,2))
ax2[0].set_title('Tile-wise predictions', fontsize='xx-large')
plt.show() # To show figure

## function to convert bbox in dictionary to geopandas
def bbox_to_geopandas(bbox_df):
    geometry = [box(x1, y1, x2, y2) for x1,y1,x2,y2 in zip(bbox_df.xmin, bbox_df.ymin, bbox_df.xmax, bbox_df.ymax)]
    poly_geo = GeoDataFrame(bbox_df, geometry=geometry)
    return poly_geo

## prepare reference and prediction bbox
### convert data types for reference bbox dictionary
reference = pd.DataFrame.from_dict(target_bbox)
reference[['xmin', 'ymin', 'xmax', 'ymax']] = reference[['xmin', 'ymin', 'xmax', 'ymax']].astype(int)

poly_reference = bbox_to_geopandas(reference)
poly_prediction_image = bbox_to_geopandas(pred_boxes)
poly_prediction_tile = bbox_to_geopandas(tile)

## settings for hvplot objects
settings_vector = dict(fill_color=None, width=400, height=400, clim=(0,1), fontsize={'title': '110%'})
settings_image = dict(x='x', y='y', data_aspect=1, xaxis=False, yaxis=None)

## create hvplot objects
plot_RGB = tc_rgb.hvplot.rgb(**settings_image, bands='channel', hover=False, rasterize=True)
plot_vector_reference = poly_reference.hvplot(hover_cols=False, legend=False).opts(title='Reference labels', alpha=1, **settings_vector)
plot_vector_image = poly_prediction_image.hvplot(hover_cols=['score'], legend=False).opts(title='Full-image predictions', alpha=0.5, **settings_vector)
plot_vector_tile = poly_prediction_tile.hvplot(hover_cols=['score'], legend=False).opts(title='Tile-wise predictions', alpha=0.5, **settings_vector)

plot_comparison = pn.Row(pn.Column(plot_RGB * plot_vector_reference, 
                         plot_RGB * plot_vector_image),
                         pn.Column(pn.Spacer(width=400, height=400),
                         plot_RGB * plot_vector_tile), scroll=True)

plot_comparison.embed()