Scribd Logo
Upload

Welcome to Scribd!

  • 5 views

    Результаты работы программы

    Uploaded by

    Влад Алексеев
    The document discusses image compression and the training of a machine learning model. It includes a table showing the peak signal-to-noise ratio (PSNR) values for 8 sample images before and after compression. It then details the training of a model over 31 epochs, recording the loss value after each epoch as it decreases with training. Finally, it mentions creating a graph of the loss for training and validation data over iterations.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content

    You might also like

    Результаты работы программы

    Uploaded by

    Влад Алексеев
    5 views3 pages
    The document discusses image compression and the training of a machine learning model. It includes a table showing the peak signal-to-noise ratio (PSNR) values for 8 sample images before and after compression. It then details the training of a model over 31 epochs, recording the loss value after each epoch as it decreases with training. Finally, it mentions creating a graph of the loss for training and validation data over iterations.

    Original Description:

    Copyright

    © © All Rights Reserved

    Available Formats

    DOCX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    The document discusses image compression and the training of a machine learning model. It includes a table showing the peak signal-to-noise ratio (PSNR) values for 8 sample images before and after compression. It then details the training of a model over 31 epochs, recording the loss value after each epoch as it decreases with training. Finally, it mentions creating a graph of the loss for training and validation data over iterations.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as docx, pdf, or txt
    5 views3 pages

    Результаты работы программы

    Uploaded by

    Влад Алексеев
    The document discusses image compression and the training of a machine learning model. It includes a table showing the peak signal-to-noise ratio (PSNR) values for 8 sample images before and after compression. It then details the training of a model over 31 epochs, recording the loss value after each epoch as it decreases with training. Finally, it mentions creating a graph of the loss for training and validation data over iterations.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as docx, pdf, or txt
    of 3

    Примеры сжатия фотографий, до и после.

    Таблица пикового отношения сигнала к шуму

    Наименование фотографии PSNR


    0001 30.602791769552557
    0002 30.657589486440312
    0003 31.815143812094227
    0004 30.857234533048338
    0005 30.181558537572307
    0006 29.417059097374413
    0007 31.099000487846947
    0008 32.13035181260656

    Средний PSNR - 30.881612890829157

    Обучение модели.
    Epoch: 1 LR: 0.00100
    Train Loss: tensor(0.0157, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 2 LR: 0.00100
    Train Loss: tensor(0.0087, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 3 LR: 0.00100
    Train Loss: tensor(0.0067, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 4 LR: 0.00100
    Train Loss: tensor(0.0056, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 5 LR: 0.00100
    Train Loss: tensor(0.0051, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 6 LR: 0.00100
    Train Loss: tensor(0.0045, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 7 LR: 0.00100
    Train Loss: tensor(0.0043, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 8 LR: 0.00100
    Train Loss: tensor(0.0040, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 9 LR: 0.00100
    Train Loss: tensor(0.0040, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 10 LR: 0.00100
    Train Loss: tensor(0.0036, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 11 LR: 0.00100
    Train Loss: tensor(0.0036, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 12 LR: 0.00100
    Train Loss: tensor(0.0034, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 13 LR: 0.00100
    Train Loss: tensor(0.0034, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 14 LR: 0.00100
    Train Loss: tensor(0.0033, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 15 LR: 0.00100
    Train Loss: tensor(0.0032, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 16 LR: 0.00100
    Train Loss: tensor(0.0031, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 17 LR: 0.00100
    Train Loss: tensor(0.0030, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 18 LR: 0.00100
    Train Loss: tensor(0.0030, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 19 LR: 0.00100
    Train Loss: tensor(0.0030, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 20 LR: 0.00100
    Train Loss: tensor(0.0029, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 21 LR: 0.00100
    Train Loss: tensor(0.0028, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 22 LR: 0.00100
    Train Loss: tensor(0.0028, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 23 LR: 0.00100
    Train Loss: tensor(0.0028, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 24 LR: 0.00100
    Train Loss: tensor(0.0027, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 25 LR: 0.00100
    Train Loss: tensor(0.0027, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 26 LR: 0.00010
    Train Loss: tensor(0.0024, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 27 LR: 0.00010
    Train Loss: tensor(0.0024, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 28 LR: 0.00010
    Train Loss: tensor(0.0024, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 29 LR: 0.00010
    Train Loss: tensor(0.0023, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 30 LR: 0.00010
    Train Loss: tensor(0.0023, device='cuda:0', grad_fn=<DivBackward0>)
    Epoch: 31 LR: 0.00010
    Train Loss: tensor(0.0023, device='cuda:0', grad_fn=<DivBackward0>)
    12284.927605867386

    График потерь при использовании обучающих данных, и неизвестных валидационных данных


    (горизонтальная – количество попыток, вертикальная – коэффициент потерь)

    You might also like