Add deep learning detection func

[CousinRock]

@giswqs Dear Professor Wu,
I have recently developed a new feature for geemap that integrates GeoAI to perform model inference directly on imagery from GEE. This functionality enables users to apply local deep learning models for instance segmentation and semantic segmentation on GEE images in an intuitive and efficient way.
I have tested this extension and it works well for various applications such as building extraction and land cover classification using NAIP imagery. I would like to contribute this feature to the geemap project, and I would greatly appreciate your feedback or suggestions on the integration and contribution process. Thank you for your time and support.

Instance segmentation

import ee
import geemap

ee.Authenticate()
ee.Initialize(project='ee-renjiewu660')

m = geemap.Map(basemap = 'SATELLITE')
m

image = ee.Image('projects/ee-renjiewu660/assets/vgee/naip_test')
dataset = ee.ImageCollection('USDA/NAIP/DOQQ')
.filter(ee.Filter.date('2017-01-01', '2018-12-31')).median().toUint8()
roi = image.geometry()
TrueColor = dataset.select(['R', 'G', 'B']).clip(roi)
TrueColorVis = {
"bands": ['R', 'G', 'B'],
"min": 0,
"max": 255,
}
m.addLayer(TrueColor, TrueColorVis, 'True Color')
m.centerObject(image)
m

model_path = "/mnt/e/DATA/TrainModel/BuildingDec/best_model.pth"

infer_area = geemap.detect_instance_segmentation(
image=TrueColor,
model_path=model_path,
scale=0.6,
window_size=512,
overlap=128,
confidence_threshold=0.5,
batch_size=4,
num_channels=3,
min_area=0,
max_tile_size=1,
max_tile_dim=512,
region = roi
)
infer_area

m.addLayer(infer_area,{},'t')
m

semantic segmentation

import ee
import geemap

ee.Authenticate()
ee.Initialize(project='ee-renjiewu660')

m = geemap.Map(basemap = 'SATELLITE')
m

image = ee.Image('projects/ee-renjiewu660/assets/vgee/m_3807511_se_18_060_20181104')
dataset = ee.ImageCollection('USDA/NAIP/DOQQ')
.filter(ee.Filter.date('2017-01-01', '2018-12-31')).median().toUint8()
roi = m.user_roi
TrueColor = dataset.clip(roi)
TrueColorVis = {
"bands": ['R', 'G', 'B'],
"min": 0,
"max": 255,
}
m.addLayer(TrueColor, TrueColorVis, 'True Color')
m.centerObject(image)
m

model_path = "/mnt/e/DATA/TrainModel/LandcoverDec/best_model.pth"

infer_area = geemap.detect_semantic_segmentation(
image=TrueColor,
model_path=model_path,
scale=0.6,
window_size=512,
overlap=128,
confidence_threshold=0.95,
batch_size=4,
num_channels=4,
num_classes=13,
min_area=0,
max_tile_size=1,
max_tile_dim=512,
region = roi
)

infer_area

m.addLayer(infer_area,{},'t')
m

[giswqs]

Thank you for your contribution. I am sorry that we won't be able to to merge this PR. The geoai package has a lot of heavy dependencies. We won't be able to add it as a dependency for geemap. You can try to contribute this PR to the geoai package.