Result Miou - Githubissues

jfzhang95 / pytorch-deeplab-xception

DeepLab v3+ model in PyTorch. Support different backbones.

MIT License

2.89k stars 779 forks source link

Result Miou #15

Closed cuishuhao closed 5 years ago

cuishuhao commented 6 years ago

could anyone show the training and testing result ? my result goes too bad to see.

emedinac commented 5 years ago

I tried some epochs but I didn't get good results. My metrics are: Loss: 281652.201770, miou: 0.3599

jfzhang95 commented 5 years ago

I think you can train model with more epoch because modified xception do not have a pretrained model and also you can use tensorboardX to visualize outputs of training data.

emedinac commented 5 years ago

well, I am training the model using the original configuration (including the pre-trained model you used) and it is in the epoch 50. I attach the current results.

[Epoch: 49, numImages: 10582]
Loss: 254079.521480
Execution time: 1499.6592651726678
Save model at .../deeplabv3plus-xception-voc_epoch-49.pth
Validation:
[Epoch: 49, numImages:  1449]
Loss: 243406.573831
MIoU: 0.387194

jfzhang95 commented 5 years ago

Did you use the latest code to train the model? Network architecture defined in previous version has some problems.

Below is my training log and results,

emedinac commented 5 years ago

Hi, ok, yes I think I am doing something wrong. because I trained another model and I got better results.

jfzhang95 commented 5 years ago

Thank you for your patience and interest in this code.

Further discussion about performance of this code can be moved to #12 .