EfficientDet.Pytorch
EfficientDet.Pytorch copied to clipboard
toandaominh1997
run demo.py, image nothing happened
Hi, thanks very much for your code! I trained a model 'checkpoint_VOC_efficientdet-d2_17.pth' on my own dataset, but when I demo.py this model no box appeared in the image. Nothing happened only image color changed.
Why need data augmentation when running demo.py? The output image changes color when I input an image.
about image color changed, you can change the demo.py code at 72 line.
#origin_img = copy.deepcopy(img)
origin_img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB) # add this code
I met the same problem when I try to test a model trained on COCO. I futher printed out the outputs of the model, and all the scores are very small numbers like 2.9548e-06 or so. Does that mean the model was not correctly trained? Although the loss seems fine after 15 epochs of training.
Hello, I'm just a newbie.
Color issue of demo.py could be fixed as https://github.com/toandaominh1997/EfficientDet.Pytorch/issues/31#issuecomment-565888531
Next, Why need data augmentation when running demo.py?
As I know data augmentation used in demo.py is
self.transform = get_augumentation(phase='test')
which is test phase, it contains some stuff like Resizing, Normalizing, ToTensor.
So it's like a preprocess? I think. (because it's test phase)
Next, nothing happened? I think model is not trained appropriately if you see "No boxes to NMS" printed
sorry for my bad English
@inFreedom92 Hi, I have met the same problem,when run eval.py to test voc2007(4952imgs), some images show "No boxes to NMS",and some other images can be detected, but finally, all classes map is <10%(0.05 or 0.06), do you know what's going on? thanks!
@pangpanghuahuamaomi
Hi, I have met the same problem,when run eval.py to test voc2007(4952imgs), some images show "No boxes to NMS",and some other images can be detected, but finally, all classes map is <10%(0.05 or 0.06), do you know what's going on? thanks!
I'm currently using my custom dataset. so I haven't used eval.py but in my opinion, the model that you're using is not trained enough...
I'm sorry that I can't give you the answer that you want.
I've got the same problem with a model trained for 100 epochs on a coco subset of 3200 images. Evaluation got mAP of 0.07 and when i run demo.py, "No boxes to NMS" message is showed. Some suggestions?
