pr_mask is not correct although the accuracy on CamVid test set is very high

qubvel-org / segmentation_models.pytorch

Semantic segmentation models with 500+ pretrained convolutional and transformer-based backbones.

https://smp.readthedocs.io/

MIT License

9.53k stars 1.66k forks source link

Closed KleinXin closed 5 years ago

KleinXin commented 5 years ago

I used se_resnext50_32x4d encoder and Unet decoder to train the segmentation model by following the CamVid tutorial on the webpage

However, the pr_mask is not correct although the accuracy on CamVid test set is very high（iou - 0.7417）

valid: 100%|████████████████████████████████████████████| 233/233 [00:10<00:00, 21.86it/s, bce_dice_loss - 0.3587, iou - 0.7417, f-score - 0.8195]

The trained model and codes can be found here

qubvel commented 5 years ago

Hi, did you use tutorial without any changes?

KleinXin commented 5 years ago

Yes, I think so.

Can you get the model and script I shared with you？

On 5/7/2019 13:26，Pavel Yakubovskiynotifications@github.com wrote：

Hi, did you use tutorial without any changes?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

qubvel commented 5 years ago

Yeah, I can. Did you try colab tutorial? No any ideas whats wrong.

crossknight commented 5 years ago

I got the same problem with other dataset using se_resnext50_32x4d encoder too.

qubvel commented 5 years ago

Line 321 bug, change model.predict -> best_model.predict

KleinXin commented 5 years ago

@qubvel problem solved, thank you very much