No Change in Descriptor Network Training Curve after 6000 Steps

[wangerniuniu]

Hello,

I am encountering an issue while training the descriptor network using dedode_detector.py. After initiating the training process, a curve is generated in the vis folder which is supposed to reflect the training progress. However, after training for 60000 steps, I've noticed that there appears to be no change in the curve, which is unexpected.

Result: The curve does not show any noticeable change after 6000 steps, suggesting that the network may not be learning as expected.

I have attached a screenshot of the curve for reference. Could you please help me understand if this behavior is normal, or if there might be an issue with the training process? Additionally, any suggestions for troubleshooting this would be greatly appreciated.

Thank you for your assistance

.

[Parskatt]

This is a bug in the plotting which should be fixed by calli plt.close. It should be clear from prints/logs that thr number of inliers go up over time.