Closed PowderYu closed 3 years ago
PowderYu
commented
3 years ago I run run.sh instead of t1_test alone. In fact, this error occurs when the program in run.sh is executed to t1_test. I am curious why there is no such error when running t1_train and t1_val. The same problem exists in tasks 2, 3, and 4. I would like to know where is my problem? thanks
JosephKJ
commented
3 years ago After parsing through the error, it seems that you are erroring out while trying to compute the likelihood. I can suggest two things: 1) are you using reliability version 0.5.6? Considering that many are able to run the code without any trouble, I suspect this to be a setup problem. 2) Can you try recreating the energy distribution parameters by running the validation step again?
Kindly reopen this if the issue persists. Thanks!
jeongeun980906
commented
3 years ago Hello. I am still having this issue even with reliability version 0.5.6. I ran validation step several times, but still doesn't work :(
PowderYu
commented
3 years ago Hello. I am still having this issue even with reliability version 0.5.6. I ran validation step several times, but still doesn't work :( I solved this problem by modifying the following function, just need to add a parameter.I think this is because of the pytorch version
def create_distribution(self, scale, shape, shift):
wd = Weibull(scale=scale, concentration=shape, validate_args=False)
transforms = AffineTransform(loc=shift, scale=1.)
weibull = TransformedDistribution(wd, transforms, validate_args=False)
return weibull
CtCCtV
commented
2 years ago Hello. I am still having this issue even with reliability version 0.5.6. I ran validation step several times, but still doesn't work :( I solved this problem by modifying the following function, just need to add a parameter.I think this is because of the pytorch version
def create_distribution(self, scale, shape, shift): wd = Weibull(scale=scale, concentration=shape, validate_args=False) transforms = AffineTransform(loc=shift, scale=1.) weibull = TransformedDistribution(wd, transforms, validate_args=False) return weibull
I can't fix the problem even though I add that parameter...Can this problem be related to the version of pytorch or detectron2? Mine are pytorch==1.8.0, detectron2==0.2.1. Thank you very much!
RuoyuChen10
commented
2 years ago Hello. I am still having this issue even with reliability version 0.5.6. I ran validation step several times, but still doesn't work :( I solved this problem by modifying the following function, just need to add a parameter.I think this is because of the pytorch version
def create_distribution(self, scale, shape, shift): wd = Weibull(scale=scale, concentration=shape, validate_args=False) transforms = AffineTransform(loc=shift, scale=1.) weibull = TransformedDistribution(wd, transforms, validate_args=False) return weibull
Thanks, that will work. Change the function in detectron2/evaluation/pascal_voc_evaluation/PascalVOCDetectionEvaluator.
Leon16226
commented
2 years ago Hello. I am still having this issue even with reliability version 0.5.6. I ran validation step several times, but still doesn't work :( I solved this problem by modifying the following function, just need to add a parameter.I think this is because of the pytorch version
def create_distribution(self, scale, shape, shift): wd = Weibull(scale=scale, concentration=shape, validate_args=False) transforms = AffineTransform(loc=shift, scale=1.) weibull = TransformedDistribution(wd, transforms, validate_args=False) return weibull
Thank you, it works. But could you example why you set this parameters?
when I run python tools/train_net.py --num-gpus 2 --eval-only --config-file ./configs/OWOD/t1/t1_test.yaml SOLVER.IMS_PER_BATCH 8 SOLVER.BASE_LR 0.005 OUTPUT_DIR "./output/t1_final" MODEL.WEIGHTS "/data/yu/code/OWOD-master/output/t1/model_final.pth"
It gave the following error：
Very much looking forward to your reply~~thanks