TypeError: cannot unpack non-iterable VisImage object

[dadizak]

thanks for this work,i'm very gratefuly to help me to fix this error:
Traceback (most recent call last):
File "/content/TextFuseNet/demo/icdar2015_detection.py", line 135, in
prediction, vis_output, polygons = detection_demo.run_on_image(img)
File "/content/TextFuseNet/demo/predictor.py", line 64, in run_on_image
vis_output, polygons = visualizer.draw_instance_predictions(predictions=instances)
TypeError: cannot unpack non-iterable VisImage object

[mwolf2]

Hi, I encountered the same error running icdar2015_detection.py. I suspect it might be something to do with the fact that I am not running TextFuseNet-Detectron due to CUDA 10.1 being incompatible with my GPU but here's the fix that is working for me right now:

In icdar2015_detection.py, change line 135 from
prediction, vis_output, polygons = detection_demo.run_on_image(img)
to
prediction, vis_output = detection_demo.run_on_image(img)
and comment out line 139.

Then, in predictor.py, change line 64 from
vis_output,polygons = visualizer.draw_instance_predictions(predictions=instances)
to
vis_output = visualizer.draw_instance_predictions(predictions=instances)
and comment out , polygons in line 66.

My guess is that their TextFuseNet-detectron's visualizer function works differently than the one in stock Detectron2. Hope this helps.

[angrylau]

pip uninstall detectron2

conda uninstall detectron2

in TextFuseNet folder：

python setup.py install

pip install fvcore-master.zip

[Hegelim]

Hi, I encountered the same error running icdar2015_detection.py. I suspect it might be something to do with the fact that I am not running TextFuseNet-Detectron due to CUDA 10.1 being incompatible with my GPU but here's the fix that is working for me right now:

In icdar2015_detection.py, change line 135 from prediction, vis_output, polygons = detection_demo.run_on_image(img) to prediction, vis_output = detection_demo.run_on_image(img) and comment out line 139.

Then, in predictor.py, change line 64 from vis_output,polygons = visualizer.draw_instance_predictions(predictions=instances) to vis_output = visualizer.draw_instance_predictions(predictions=instances) and comment out , polygons in line 66.

My guess is that their TextFuseNet-detectron's visualizer function works differently than the one in stock Detectron2. Hope this helps.

There would still be problems in save_result_to_txt as it takes in polygons as an argument