1700 stop

[JX-Github-web]

have you solved this problem yet?

[stan1512]

train.py 135
if iteration % 100 == 0:
if iteration > opt.min_iters and ema_depthloss_for_log > prev_depthloss:
training_report(tb_writer, iteration, Ll1, loss, l1_loss, iter_start.elapsed_time(iter_end), [iteration], scene, render,
(pipe, background), txt_path = os.path.join(args.model_path, "metric.txt"))
scene.save(iteration)
print(f"!!! Stop Point: {iteration} !!!")
break
else:
prev_depthloss = ema_depthloss_for_log
eliminate this early end logic

[JX-Github-web]

如果迭代 % 100 == 0，则 train.py 135：如果迭代 > opt.min_iters 且ema_depthloss_for_log > prev_depthloss：training_report（tb_writer， iteration， Ll1， loss， l1_loss， iter_start.elapsed_time（iter_end）， [iteration]， scene， render， （pipe， background）， txt_path = os.path.join（args.model_path， “metric.txt”）） scene.save（iteration） print（f“!!停止点： {iteration} !!”） break else：prev_depthloss = ema_depthloss_for_log 消除此早期结束逻辑

Thanks, this one works