Text Prompt Segmentation¶

This example demonstrates how to use the geoai-py package for text prompt segmentation using CLIPSeg.

Install package¶

To use the geoai-py package, ensure it is installed in your environment. Uncomment the command below if needed.

In [1]:

# %pip install geoai-py

# %pip install geoai-py

Import libraries¶

In [2]:

import geoai

import geoai

Download sample data¶

In [3]:

raster_url = (
    "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/trees_brazil.tif"
)
raster_path = geoai.download_file(raster_url)

raster_url = ( "https://huggingface.co/datasets/giswqs/geospatial/resolve/main/trees_brazil.tif" ) raster_path = geoai.download_file(raster_url)

trees_brazil.tif:   0%|          | 0.00/3.93M [00:00<?, ?B/s]

trees_brazil.tif: 100%|██████████| 3.93M/3.93M [00:00<00:00, 97.8MB/s]

Visualize data¶

In [4]:

geoai.view_raster(raster_url)

geoai.view_raster(raster_url)

Out[4]:

Initialize model¶

In [5]:

segmenter = geoai.CLIPSegmentation(tile_size=512, overlap=32)

segmenter = geoai.CLIPSegmentation(tile_size=512, overlap=32)

Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

WARNING:transformers.processing_utils:Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.52, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`.

Model loaded on cpu

Run inference¶

In [6]:

output_path = "tree_masks.tif"
text_prompt = "trees"

output_path = "tree_masks.tif" text_prompt = "trees"

In [7]:

segmenter.segment_image(
    raster_path,
    output_path=output_path,
    text_prompt=text_prompt,
    threshold=0.5,
    smoothing_sigma=1.0,
)

segmenter.segment_image( raster_path, output_path=output_path, text_prompt=text_prompt, threshold=0.5, smoothing_sigma=1.0, )

Processing tiles:   0%|          | 0/15 [00:00<?, ?it/s]

Processing tiles:   7%|▋         | 1/15 [00:00<00:10,  1.37it/s]

Processing tiles:  13%|█▎        | 2/15 [00:01<00:09,  1.41it/s]

Processing tiles:  20%|██        | 3/15 [00:02<00:08,  1.46it/s]

Processing tiles:  27%|██▋       | 4/15 [00:02<00:07,  1.46it/s]

Processing tiles:  33%|███▎      | 5/15 [00:03<00:06,  1.46it/s]

Processing tiles:  40%|████      | 6/15 [00:04<00:06,  1.45it/s]

Processing tiles:  47%|████▋     | 7/15 [00:04<00:05,  1.48it/s]

Processing tiles:  53%|█████▎    | 8/15 [00:05<00:04,  1.47it/s]

Processing tiles:  60%|██████    | 9/15 [00:06<00:04,  1.48it/s]

Processing tiles:  67%|██████▋   | 10/15 [00:06<00:03,  1.51it/s]

Processing tiles:  73%|███████▎  | 11/15 [00:07<00:02,  1.52it/s]

Processing tiles:  80%|████████  | 12/15 [00:08<00:01,  1.53it/s]

Processing tiles:  87%|████████▋ | 13/15 [00:08<00:01,  1.53it/s]

Processing tiles:  93%|█████████▎| 14/15 [00:09<00:00,  1.51it/s]

Processing tiles: 100%|██████████| 15/15 [00:10<00:00,  1.52it/s]

Processing tiles: 100%|██████████| 15/15 [00:10<00:00,  1.49it/s]

Segmentation saved to tree_masks.tif

Out[7]:

'tree_masks.tif'

Visualize results¶

In [8]:

geoai.view_raster(
    output_path,
    nodata=0,
    opacity=0.8,
    colormap="greens",
    layer_name="Trees",
    basemap=raster_url,
)

geoai.view_raster( output_path, nodata=0, opacity=0.8, colormap="greens", layer_name="Trees", basemap=raster_url, )

Out[8]:

In [9]:

geoai.create_split_map(
    left_layer=output_path,
    right_layer=raster_url,
    left_label="Trees",
    right_label="Satellite Image",
    left_args={"nodata": 0, "opacity": 0.8, "colormap": "greens"},
    basemap=raster_url,
)

geoai.create_split_map( left_layer=output_path, right_layer=raster_url, left_label="Trees", right_label="Satellite Image", left_args={"nodata": 0, "opacity": 0.8, "colormap": "greens"}, basemap=raster_url, )

Out[9]: