step1： synthetic_shapes_v6

[yyyyiiiinnnn]

Hello, I found that there are 10,000 pictures in the train file used for training in the synthetic_shapes_v6 generated after running the first step, but my GPU seems to be unable to afford it. How should I only include 5,000 pictures for my subsequent training? Please help me with some confusion. Thanks!

[rpautrat]

Hi, in superpoint/configs/magic-point_shapes.yaml, you can add the following entry:

data:
    generation:
        split_sizes:
            training: 5000

But I don't understand why 10000 would be too much for your GPU... The whole set of images is never at the same time in GPU memory, it should only be generated and processed by batches only.

[yyyyiiiinnnn]

My friend!Thank you very much for your answer!

[yyyyiiiinnnn]

I'm so sorry, I'm really new to this, I can't quite figure out the process of it, I set my batch count to 1.
The reason I ask this question is because I also get the error mentioned in #296 :
Start training
[07/27/2023 17:23:31 INFO] Iter 0: loss 4.1785, precision 0.0006, recall 0.0545
/home/yibo/SuperPoint-master/superpoint/models/base_model.py:387: RuntimeWarning: Mean of empty slice
metrics = {m: np.nanmean(metrics[m], axis=0) for m in metrics}
[07/27/2023 17:23:44 INFO] Iter 1000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:23:57 INFO] Iter 2000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:24:10 INFO] Iter 3000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:24:23 INFO] Iter 4000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:24:36 INFO] Iter 5000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:24:49 INFO] Iter 6000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:25:02 INFO] Iter 7000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:25:15 INFO] Iter 8000: loss nan, precision nan, recall 0.0000
[07/27/2023 17:25:28 INFO] Iter 9000: loss nan, precision nan, recall 0.0000

[1z2213]

很抱歉，我真的很陌生，我不太清楚它的过程，我将我的批次计数设置为 1。我问这个问题的原因是因为我也收到了 #296 中提到的错误： 开始训练 [07/27/2023 17：23：31 INFO] Iter 0： 损失 4.1785，精度 0.0006，召回 0.0545 /home/yibo/SuperPoint-master/superpoint/models/base_model.py：387： 运行时警告：空切片指标的平均值 = {m： np.nanmean（metrics[m]， axis=0） for m in metrics} [07/27/2023 17：23：44 INFO] Iter 1000： 损失楠，精密楠，召回 0.0000 [07/27/2023 17：23：57 信息] Iter 2000：损失楠，精密楠，召回 0.0000 [07/27/2023 17：24：10 信息] Iter 3000：损失楠，精密楠，召回 0.0000 [07/27/2023 17：24：23 信息] Iter 4000：损失楠，精度楠，召回 0.0000 [07/27/2023 17：24：36 信息] Iter 5000：损失楠，精密楠，召回0.0000 [07/27/2023 17：24：49 INFO] Iter 6000：损失楠， 精度楠，召回 0.0000 [07/27/2023 17：25：02 信息] Iter 7000：损失楠，精密楠，召回 0.0000 [07/27/2023 17：25：15 信息] Iter 8000：损失楠，精密楠，召回 0.0000 [07/27/2023 17：25：28 信息] Iter 9000：损失楠，精密楠，召回 0.0000

HI! Have you solved this problem yet?

[yyyyiiiinnnn]

For me, the reason for this problem seems to be that I used my own pictures for training. When I use coco to run step1, this error is solved.