There is no scheduler at saving checkpoint

Hyungson commented 4 months ago

hi, i tried to resume train srgan, but there is no scheduler at saving checkpoint. Is this my issue, or code issue?


save_checkpoint({"epoch": epoch + 1,
                         "psnr": psnr,
                         "ssim": ssim,
                         "state_dict": g_model.state_dict(),
                         "ema_state_dict": ema_g_model.state_dict() if ema_g_model is not None else None,
                         "optimizer": g_optimizer.state_dict()},
                        f"g_epoch_{epoch + 1}.pth.tar",
                        samples_dir,
                        results_dir,
                        "g_best.pth.tar",
                        "g_last.pth.tar",
                        is_best,
                        is_last)
        save_checkpoint({"epoch": epoch + 1,
                         "psnr": psnr,
                         "ssim": ssim,
                         "state_dict": d_model.state_dict(),
                         "optimizer": d_optimizer.state_dict()},
                        f"d_epoch_{epoch + 1}.pth.tar",
                        samples_dir,
                        results_dir,
                        "d_best.pth.tar",
                        "d_last.pth.tar",
                        is_best,
                        is_last)

File "/home/work/dacon/srgan/train_gan.py", line 98, in main g_model, ema_g_model, start_epoch, best_psnr, best_ssim, g_optimizer, g_scheduler = load_resume_state_dict( File "/home/work/dacon/srgan/utils.py", line 165, in load_resume_state_dict scheduler.load_state_dict(checkpoint["scheduler"]) KeyError: 'scheduler'

nanakoMI commented 2 months ago

我也遇到了同样的问题

Hyungson commented 2 months ago

我也遇到了同样的问题 save_checkpoint({"epoch": epoch + 1, "psnr": psnr, "ssim": ssim, "state_dict": g_model.state_dict(), "ema_state_dict": ema_g_model.state_dict() if ema_g_model is not None else None, "optimizer": g_optimizer.state_dict(), 'scheduler': g_scheduler.state_dict()}, f"gepoch{epoch + 1}.pth.tar", samples_dir, results_dir, "g_best.pth.tar", "g_last.pth.tar", is_best, is_last) save_checkpoint({"epoch": epoch + 1, "psnr": psnr, "ssim": ssim, "state_dict": d_model.state_dict(), "optimizer": d_optimizer.state_dict(), 'scheduler': d_scheduler.state_dict()}, f"depoch{epoch + 1}.pth.tar", samples_dir, results_dir, "d_best.pth.tar", "d_last.pth.tar", is_best, is_last)

i added state_dict

nanakoMI commented 2 months ago

谢谢你的回复!

我已经解决了这个问题，主要原因是预训练模型的路径存在错误。

期待下次交流！

best wishes！

--

Computer Vision Research China Agricultural University Mina Han Mobile. 13103573112

在 2024-06-11 07:22:31，"임형빈(LYM-HYEUNGBIN)" @.***> 写道：

我也遇到了同样的问题

save_checkpoint({"epoch": epoch + 1, "psnr": psnr, "ssim": ssim, "state_dict": g_model.state_dict(), "ema_state_dict": ema_g_model.state_dict() if ema_g_model is not None else None, "optimizer": g_optimizer.state_dict(), 'scheduler': g_scheduler.state_dict()}, f"gepoch{epoch + 1}.pth.tar", samples_dir, results_dir, "g_best.pth.tar", "g_last.pth.tar", is_best, is_last) save_checkpoint({"epoch": epoch + 1, "psnr": psnr, "ssim": ssim, "state_dict": d_model.state_dict(), "optimizer": d_optimizer.state_dict(), 'scheduler': d_scheduler.state_dict()},

                f"d_epoch_{epoch + 1}.pth.tar",
                samples_dir,
                results_dir,
                "d_best.pth.tar",
                "d_last.pth.tar",
                is_best,
                is_last)

i added state_dict

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

Lornatang / SRGAN-PyTorch

There is no scheduler at saving checkpoint #92