sam.predict with BBOX : No valid pixel coordinates found. TypeError: 'NoneType' object is not subscriptable

[alex6H]

Hello,

First of all thanks a lot for this great lib !

I have an issue with SAM and the bbox segmentation. I followed the tutorial here which run perfectly on my notebook. When I use my data which are from a .shp I have a error "No valid pixel coordinates found."
TypeError: 'NoneType' object is not subscriptable

Here is my code :

# Initiate
sam = SamGeo(
    model_type="vit_b",
    automatic=False,
    device='cuda',
    sam_kwargs=None,
)

# Load Geotiff
sam.set_image(SOURCE_IMAGE_PATH)

# Read .shp
gdf_mask = gpd.read_file("output.shp").to_crs(4326)

# Create bbox
gdf_bboxes = gdf_mask['geometry'].bounds
bbox_list = gdf_bboxes[['minx','miny','maxx','maxy']].to_numpy().tolist()
bbox_list


# Use box prompts
sam.predict(
            boxes= bbox_list,
            point_crs="EPSG:4326",
            output="mask2.tif", 
            dtype="uint8",
           )
)

Here is the error:

No valid pixel coordinates found.

---

TypeError Traceback (most recent call last)
Cell In[78], line 2
1 # Use box prompts
----> 2 sam.predict(
3 boxes= bbox_list,
4 point_crs="EPSG:3857",
5 output="mask2.tif",
6 dtype="uint8",
7 )

File ~\Anaconda3\envs\Yolo8\lib\site-packages\samgeo\samgeo.py:559, in SamGeo.predict(self, point_coords, point_labels, boxes, point_crs, mask_input, multimask_output, return_logits, output, index, mask_multiplier, dtype, return_results, **kwargs)
557 coords = bbox_to_xy(self.source, boxes, point_crs)
558 input_boxes = np.array(coords)
--> 559 if isinstance(coords[0], int):
560 input_boxes = input_boxes[None, :]
561 else:

TypeError: 'NoneType' object is not subscriptable

My bbox_list look like this:
[[29.12381803879782,
-1.611772506698625,
29.124026117545267,
-1.611600683403621],
[29.123034288297987,
-1.6135127229398778,
29.12307628216605,
-1.6134663678379761]]

I have tried different format but can't find a way to have a segmentation.

Can you help me ?

Thanks !

[giswqs]

What is the crs of the image?

[alex6H]

Also EPSG:4326

When I open both (.shp and .tif) in QGIS they do overlap and ESPG are the same

[alex6H]

I forgot to mention that my .tif is not a RGB but an orthomosaic create from a set of .jpg (from drone images)

[giswqs]

Try to reproject the image so that it has a projection, such as EPSG:3857

[alex6H]

You are my hero !

Solution :
raster in EPSG:3857
.shp stays in EPSG:4326

then :

# Use box prompts
sam.predict(
            boxes= bbox_list,
            point_crs="EPSG:4326",
            output="mask2.tif", 
            dtype="uint8",
           )

[alex6H]

Thank you so much !

[giswqs]

Glad to hear that it worked.