detect-objects

Name: detect-objects
Author: opengeos

23stars2forks

How to install detect-objects?

opengeos/geoai-skills/detect-objects

$npx -y skills add opengeos/geoai-skills --skill detect-objects

Installs into the current project.

Summary

Run pre-trained AI models on geospatial imagery. Detect buildings, cars, ships, solar panels, agriculture fields, or use text-prompted segmentation with GroundedSAM. Requires GPU for best performance.

Files · 1

View on GitHub

SKILL.md

1	You are helping the user run AI object detection on geospatial imagery using geoai.
2
3	Input: `$@`
4
5	Follow these steps in order.
6
7	## Step 1 -- Parse arguments
8
9	Extract:
10	- `$0` as the model name: `buildings`, `cars`, `ships`, `solar-panels`, `parking-lots`, `agriculture`, or `grounded-sam`
11	- `$1` as the input raster path
12	- `--text PROMPT` for GroundedSAM text-prompted segmentation (required when model is `grounded-sam`)
13	- `--output FILE` for the output vector file (default: `./<model>_detections.gpkg`)
14
15	If the model name is not recognized, list the available models and ask the user to pick one.
16
17	Model mapping:
18
19	\| Argument \| GeoAI Class \|
20	\|---\|---\|
21	\| `buildings` \| `geoai.BuildingFootprintExtractor` \|
22	\| `cars` \| `geoai.CarDetector` \|
23	\| `ships` \| `geoai.ShipDetector` \|
24	\| `solar-panels` \| `geoai.SolarPanelDetector` \|
25	\| `parking-lots` \| `geoai.ParkingSplotDetector` \|
26	\| `agriculture` \| `geoai.AgricultureFieldDelineator` \|
27	\| `grounded-sam` \| `geoai.GroundedSAM` \|
28
29	## Step 2 -- Check GPU availability
30
31	```bash
32	python3 -c "
33	import torch
34	if torch.cuda.is_available():
35	print(f'GPU: {torch.cuda.get_device_name(0)}')
36	print(f'CUDA: {torch.version.cuda}')
37	print(f'Memory: {torch.cuda.get_device_properties(0).total_mem / 1e9:.1f} GB')
38	else:
39	print('GPU: not available (CPU mode)')
40	print('Warning: inference will be significantly slower without a GPU')
41	"
42	```
43
44	If no GPU is available, warn the user but continue.
45
46	## Step 3 -- Resolve the input file
47
48	If `$1` looks like an absolute path, use it directly. Otherwise:
49
50	```bash
51	find "$PWD" -name "$1" -not -path '/.git/' 2>/dev/null
52	```
53
54	If no file specified and state exists, check for recently inspected/downloaded files:
55
56	```bash
57	STATE_DIR=""
58	test -f .geoai-skills/state.json && STATE_DIR=".geoai-skills"
59	PROJECT_ROOT="$(git rev-parse --show-toplevel 2>/dev/null \|\| echo "$PWD")"
60	PROJECT_ID="$(echo "$PROJECT_ROOT" \| tr '/' '-')"
61	test -f "$HOME/.geoai-skills/$PROJECT_ID/state.json" && STATE_DIR="$HOME/.geoai-skills/$PROJECT_ID"
62	```
63
64	## Step 4 -- Run the detector
65
66	### Pre-trained detectors (buildings, cars, ships, solar-panels, parking-lots, agriculture)
67
68	```bash
69	python3 -c "
70	import geoai
71
72	detector = geoai.DETECTOR_CLASS()
73	gdf = detector.predict(
74	'INPUT_PATH',
75	output_path='OUTPUT_PATH',
76	)
77	print(f'Detections: {len(gdf)}')
78	print(f'Output: OUTPUT_PATH')
79	print(f'Columns: {list(gdf.columns)}')
80	if len(gdf) > 0:
81	print('---')
82	print('Sample (first 5):')
83	print(gdf.head().to_string())
84	"
85	```
86
87	Replace `DETECTOR_CLASS` with the appropriate class from the mapping table (e.g. `BuildingFootprintExtractor`).
88
89	### GroundedSAM (text-prompted segmentation)
90
91	```bash
92	python3 -c "
93	import geoai
94
95	sam = geoai.GroundedSAM()
96	gdf = sam.predict(
97	'INPUT_PATH',
98	text_prompt='TEXT_PROMPT',
99	output_path='OUTPUT_PATH',
100	)
101	print(f'Segments: {len(gdf)}')
102	print(f'Output: OUTPUT_PATH')
103	print(f'Columns: {list(gdf.columns)}')
104	if len(gdf) > 0:
105	print('---')
106	print('Sample (first 5):')
107	print(gdf.head().to_string())
108	"
109	```
110
111	Replace `TEXT_PROMPT` with the user's text prompt.
112
113	Replace `INPUT_PATH` and `OUTPUT_PATH` with actual values before running.
114
115	## Step 5 -- Report results
116
117	Summarize:
118	- Model used
119	- Number of detections/segments
120	- Output file path
121	- Sample of results
122
123	Then suggest: "Use `/geoai-skills:inspect-geo` to examine the detection output."
124
125	## Error handling
126
127	- `import geoai` fails -> delegate to `/geoai-skills:install-geoai`.
128	- `import torch` fails -> suggest installing PyTorch: `pip install torch torchvision`.
129	- CUDA out of memory -> suggest reducing the tile size or processing a smaller area. If the detector accepts a `tile_size` parameter, recommend a smaller value.
130	- Model download fails -> check network connectivity. Models are downloaded from Hugging Face on first use.
131	- Input is not a raster -> suggest using a GeoTIFF file. If the user has a vector file, suggest `/geoai-skills:process-raster vector-to-raster` first.
132	- GroundedSAM without --text -> ask the user for a text prompt describing what to detect.

People who liked this also viewed

6 picks

Type

detect-objects

23stars2forks

View on GitHub

How to install detect-objects?

opengeos/geoai-skills/detect-objects

$npx -y skills add opengeos/geoai-skills --skill detect-objects

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	You are helping the user run AI object detection on geospatial imagery using geoai.
2
3	Input: `$@`
4
5	Follow these steps in order.
6
7	## Step 1 -- Parse arguments
8
9	Extract:
10	- `$0` as the model name: `buildings`, `cars`, `ships`, `solar-panels`, `parking-lots`, `agriculture`, or `grounded-sam`
11	- `$1` as the input raster path
12	- `--text PROMPT` for GroundedSAM text-prompted segmentation (required when model is `grounded-sam`)
13	- `--output FILE` for the output vector file (default: `./<model>_detections.gpkg`)
14
15	If the model name is not recognized, list the available models and ask the user to pick one.
16
17	Model mapping:
18
19	\| Argument \| GeoAI Class \|
20	\|---\|---\|
21	\| `buildings` \| `geoai.BuildingFootprintExtractor` \|
22	\| `cars` \| `geoai.CarDetector` \|
23	\| `ships` \| `geoai.ShipDetector` \|
24	\| `solar-panels` \| `geoai.SolarPanelDetector` \|
25	\| `parking-lots` \| `geoai.ParkingSplotDetector` \|
26	\| `agriculture` \| `geoai.AgricultureFieldDelineator` \|
27	\| `grounded-sam` \| `geoai.GroundedSAM` \|
28
29	## Step 2 -- Check GPU availability
30
31	```bash
32	python3 -c "
33	import torch
34	if torch.cuda.is_available():
35	print(f'GPU: {torch.cuda.get_device_name(0)}')
36	print(f'CUDA: {torch.version.cuda}')
37	print(f'Memory: {torch.cuda.get_device_properties(0).total_mem / 1e9:.1f} GB')
38	else:
39	print('GPU: not available (CPU mode)')
40	print('Warning: inference will be significantly slower without a GPU')
41	"
42	```
43
44	If no GPU is available, warn the user but continue.
45
46	## Step 3 -- Resolve the input file
47
48	If `$1` looks like an absolute path, use it directly. Otherwise:
49
50	```bash
51	find "$PWD" -name "$1" -not -path '/.git/' 2>/dev/null
52	```
53
54	If no file specified and state exists, check for recently inspected/downloaded files:
55
56	```bash
57	STATE_DIR=""
58	test -f .geoai-skills/state.json && STATE_DIR=".geoai-skills"
59	PROJECT_ROOT="$(git rev-parse --show-toplevel 2>/dev/null \|\| echo "$PWD")"
60	PROJECT_ID="$(echo "$PROJECT_ROOT" \| tr '/' '-')"
61	test -f "$HOME/.geoai-skills/$PROJECT_ID/state.json" && STATE_DIR="$HOME/.geoai-skills/$PROJECT_ID"
62	```
63
64	## Step 4 -- Run the detector
65
66	### Pre-trained detectors (buildings, cars, ships, solar-panels, parking-lots, agriculture)
67
68	```bash
69	python3 -c "
70	import geoai
71
72	detector = geoai.DETECTOR_CLASS()
73	gdf = detector.predict(
74	'INPUT_PATH',
75	output_path='OUTPUT_PATH',
76	)
77	print(f'Detections: {len(gdf)}')
78	print(f'Output: OUTPUT_PATH')
79	print(f'Columns: {list(gdf.columns)}')
80	if len(gdf) > 0:
81	print('---')
82	print('Sample (first 5):')
83	print(gdf.head().to_string())
84	"
85	```
86
87	Replace `DETECTOR_CLASS` with the appropriate class from the mapping table (e.g. `BuildingFootprintExtractor`).
88
89	### GroundedSAM (text-prompted segmentation)
90
91	```bash
92	python3 -c "
93	import geoai
94
95	sam = geoai.GroundedSAM()
96	gdf = sam.predict(
97	'INPUT_PATH',
98	text_prompt='TEXT_PROMPT',
99	output_path='OUTPUT_PATH',
100	)
101	print(f'Segments: {len(gdf)}')
102	print(f'Output: OUTPUT_PATH')
103	print(f'Columns: {list(gdf.columns)}')
104	if len(gdf) > 0:
105	print('---')
106	print('Sample (first 5):')
107	print(gdf.head().to_string())
108	"
109	```
110
111	Replace `TEXT_PROMPT` with the user's text prompt.
112
113	Replace `INPUT_PATH` and `OUTPUT_PATH` with actual values before running.
114
115	## Step 5 -- Report results
116
117	Summarize:
118	- Model used
119	- Number of detections/segments
120	- Output file path
121	- Sample of results
122
123	Then suggest: "Use `/geoai-skills:inspect-geo` to examine the detection output."
124
125	## Error handling
126
127	- `import geoai` fails -> delegate to `/geoai-skills:install-geoai`.
128	- `import torch` fails -> suggest installing PyTorch: `pip install torch torchvision`.
129	- CUDA out of memory -> suggest reducing the tile size or processing a smaller area. If the detector accepts a `tile_size` parameter, recommend a smaller value.
130	- Model download fails -> check network connectivity. Models are downloaded from Hugging Face on first use.
131	- Input is not a raster -> suggest using a GeoTIFF file. If the user has a vector file, suggest `/geoai-skills:process-raster vector-to-raster` first.
132	- GroundedSAM without --text -> ask the user for a text prompt describing what to detect.

People who liked this also viewed

6 picks

Type