yxl23

Results 40 comments of yxl23

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

My epochs are 800, and I tried to turn off data augmentation, but the problem still persists

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

Author, can you provide me with a complete OBB training process, including dataset preparation, format conversion, and parameter settings for the training section

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

And when I finished training and went to verify, there was nothing on the images in the validation set, and all the indicators were 0 ![image](https://github.com/user-attachments/assets/fa0c8ac0-b03d-4b4d-873a-93abcbefed0b)

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

Hello author, I wonder if it's because there are more data with angles of 0 in my dataset than those with angles, which leads to the problem of detection box...

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

These are my training parameters, and I still have this problem. Can you help me check if any switch was not turned on during my training? The results after training...

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

Yes, including the annotation of val_match.0.jpg, it is also correct, that is, val_pred.0.jpg is incorrect

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

Author, I have no problem annotating the data, but is it possible that my data did not split high-resolution images into low resolution ones, causing the detection box to shift

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

My version is already the latest, and I have tried cutting it down to low resolution, but the problem still hasn't been solved

Hello author, while using OBB to train my dataset, I found that the tilted boxes in the validation set images and training set images after training did not match. There were cases where the angles were skewed, and during my inference, the detection boxes were completely different from what I labeled

![val_batch0_pred](https://github.com/user-attachments/assets/31ec1731-6973-457f-b714-35172deee6b9) ![results](https://github.com/user-attachmen ![train_batch492](https://github.com/ ![train_batch2](https://github.com/user-attachments/assets/11ee049f-c367-4e1a-b021-7a6433bf0c02) user-attachments/assets/a4e31251-0167-4 ![train_batch1](https://github.com/user-attachments/assets/87f2b ![train_batch0](https://github.com/user-attachments/assets/805b1afb-56a9-4c9c-89cf-c4934da4ceb9) daa-7d0e-48b3-a4e8-65b0d9387722) c88-947d-95777ff0784e) ts/assets/ced2bfd8-ba4a-4271-9e46-0 ![train_batch490](https://github.com/user-attachments/assets/d9a57eb7-caf0-4170-aeef-ba678f53da6a) ![train_batch491](https://github.com/user-attachments/assets/e93882ca-796f-43db-805c-3eb2e5ac39dc) 61160b2f3ec) ![val_batch0_labels](https://github.com/user-attachments/assets/417bc83c-0014-4baa-8c4a-114b9322827f)

`Trainer.fit` stopped: `max_epochs=1` reached. Epoch 0/0 ------------------ 2/2 0:00:15 • 0:00:00 6.37it/s pixel_AUROC: 0.000 pixel_F1Score: 0.000

My mask is correct, but it is very small because my industrial defects are very small