loss大概多少，分割效果才说得过去

[songhuaqinggg]

自己训练loss下降非常慢，达到0.0014分割的还是不准，想知道大家训练到loss多少，分割的才算比较准

[theneao]

据说其他人要10w个step才差不多

[DWBSIC]

据说其他人要10w个step才差不多

我直接20w个step，效果还可以

[songhuaqinggg]

请问您最后loss大概多少哈，还有跑了多少个epoch，万分感谢~

[cannonli7]

请问一下，训练的run_loop函数是不是按照默认参数终止不了？

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Excuse me, can the training run_loop not be terminated according to the default parameters?

[JAYCHOU2020]

自己训练loss下降非常慢，达到0.0014分割的还是不准，想知道大家训练到loss多少，分割的才算比较准

你这个loss是在达到多少step得到的呢？

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

The loss of self-training decreases very slowly, and the segmentation is still inaccurate when it reaches 0.0014. I would like to know how much loss you have trained to, so that the segmentation is more accurate.

In how many steps did you achieve this loss?

[princerice]

据说其他人要10w个step才差不多

我直接20w个step，效果还可以

想问下你的显存和batch多大呢 20w步得多久啊

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

It is said that other people need 100,000 steps to get it close.

I just did 200,000 steps, and the effect was okay

I would like to ask how big your video memory and batch size are. How long does it take to get 200,000 steps?

[princerice]

据说其他人要10w个step才差不多

我直接20w个step，效果还可以

How big is your memory and batch? How long will 20wstep take