Keras-retinanet: "TypeError: Expected Ptr<cv::UMat> for argument 'img' "while running retinanet-debug on custom csv file.

[Sehjbir]

File "/opt/conda/bin/retinanet-debug", line 8, in
sys.exit(main())
File "/opt/conda/lib/python3.7/site-packages/keras_retinanet/bin/debug.py", line 321, in main
run(generator, args, anchor_params=anchor_params, pyramid_levels=pyramid_levels)
File "/opt/conda/lib/python3.7/site-packages/keras_retinanet/bin/debug.py", line 228, in run
draw_annotations(image, annotations, color=(0, 0, 255), label_to_name=generator.label_to_name)
File "/opt/conda/lib/python3.7/site-packages/keras_retinanet/utils/visualization.py", line 105, in draw_annotations
draw_caption(image, annotations['bboxes'][i], caption)
File "/opt/conda/lib/python3.7/site-packages/keras_retinanet/utils/visualization.py", line 45, in draw_caption
cv2.putText(image, caption, (b[0], b[1] - 10), cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 2)
TypeError: Expected Ptr for argument 'img'

[Sehjbir]

command run:
retinanet-debug --no-resize --show-annotations --no-gui --output-dir.... --> TypeError: Expected Ptrcv::UMat for argument 'img'
retinanet-debug --show-annotations --no-gui --output-dir.... --> No Error

-no-resize flag along with --show-annotations is throwing this error

[udaylunawat]

Try converting to image format from array using

from PIL import Image

Image.fromarray(image)

[TimNagle-McNaughton]

Getting the same error when running evaluate. Images are 100x100 greyscale.

[joluga24]

I'm getting the same error with evaluate.py , did someone solve the issue yet?

[TimNagle-McNaughton]

For greyscale, this worked for me:

At line 78:

raw_image = generator.load_image(i)
image = np.asarray(raw_image) (add this line)

[Sehjbir]

I added the line image = np.asarray(image) in debug.py file. However, I'm getting the same error

[guidotrucco]

In _debug.py_, I fixed this by adding:
image = np.float32(image)
in line 204, after generator.load_image(i)

[TimNagle-McNaughton]

@Sehjbir you need to reference "raw_image" not "image", @guidotrucco 's answer works too

[Sehjbir]

@TimNagle-McNaughton yes thanks. I tried @guidotrucco solution. I can see the annotations now for --no-resize flag. Thanks :)

[juanmanuellopezcorrea]

Hello Everyone, you know where I have to changes in evaluate.py the solution @guidotrucco

[TimNagle-McNaughton]

As he said:

"image = np.float32(image)
in line 204, after generator.load_image(i)"

Unless I'm missing something?

[juanmanuellopezcorrea]

Hello TimNagle. Thanks for your answer. My problem is evaluate.py when I run keras-retinanet/keras_retinanet/bin/evaluate.py --gpu 0 --score-threshold 0.001 --convert-model --image-min-side 1296 --save-path "/Backup/home/jmlopez/retinanet/prediction/" --image-max-side 3000 csv "/home/jmlopez/RetinaNet/keras-retinanet/train_set.csv" "/home/jmlopez/RetinaNet/keras-retinanet/classes.csv" "/Backup/home/jmlopez/retinanet/snapshots/resnet50_csv_90.h5" > results/test_90_001.txt

[TimNagle-McNaughton]

So are you having the same problem but when you run evaluate? If so I would make a new issue.

[juanmanuellopezcorrea]

No, it is about this issue because but in evaluate.py I call on evaluate.py -save-path and i got this error :
File "keras-retinanet/keras_retinanet/bin/../../keras_retinanet/utils/visualization.py", line 45, in draw_caption
cv2.putText(image, caption, (b[0], b[1] - 10), cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 2)
TypeError: Expected Ptr for argument 'img'
Then
white that image = np.float32(image) I solved this error but a the images don't have a bounding box and any annotations