Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Depthwise Iyapa Convolutions adanwo

Apply the depthwise and pointwise split, compare convolution costs, calculate a small example, and recognize efficient mobile-model design.

Jẹmọ itọsọna awọn ọnaDepthwise Iyapa Convolutions
Ibeere 1 ti 8

Kini awọn igbesẹ meji ti ariyanjiyan ti o ya sọtọ ti o jinlẹ ti o pin ipinya boṣewa kan si?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini awọn igbesẹ meji ti ariyanjiyan ti o ya sọtọ ti o jinlẹ ti o pin ipinya boṣewa kan si?
    • Igbesẹ ikojọpọ atẹle nipa igbesẹ deede
    • Awọn iyipo 3x3 tolera meji pẹlu ReLU laarin wọn
    • Àlẹmọ aye ijinle ti o jinlẹ fun ikanni kan, lẹhinna adapọ ikanni 1x1 wise pointwise
    • Ayipada kooduopo ati deconvolution decoder
  2. Ni igbesẹ ti o jinlẹ, bawo ni a ṣe tọju awọn ikanni titẹ sii?
    • Ikanni kọọkan jẹ filtered ni ominira pẹlu ekuro tirẹ
    • Gbogbo awọn ikanni ti wa ni akopọ ni akọkọ
    • Awọn ikanni ti wa ni laileto dapọ ṣaaju ṣiṣe sisẹ
    • Nikan ikanni akọkọ ti wa ni convolved
  3. For many output channels, what operation-count reduction does the guide associate with a 3×3 depthwise separable convolution?
    • Nipa awọn akoko 2 kere si
    • Nipa 100 igba kere
    • O nlo iṣiro diẹ sii, kii ṣe kere si
    • Nipa awọn akoko 8 si 9 kere si
  4. Kini ipa ti itọka-ọrọ (1x1) ni apẹrẹ yii?
    • Lati downsample ipinnu aaye
    • Lati darapo alaye kọja awọn ikanni ni ipo kọọkan
    • Lati fi awọn ifaminsi ipo kun
    • Lati yọ awọn ofin abosi kuro
  5. Iru faaji ti a mọ daradara wo ni o jẹ ki awọn iyipada iyapa jinlẹ jinlẹ fun awọn ẹrọ alagbeka?
    • MobileNet
    • ResNet
    • AlexNet
    • LSTM
  6. An implementation applies only the depthwise spatial filters and omits the pointwise layer. Which capability is missing from that block?
    • Spatial filtering within each channel
    • Separate kernels for the input channels
    • Learned combination of information across channels
    • Local spatial feature extraction
  7. Using the guide’s formulas with M = 4 input channels and N = 16 outputs, how many multiply-adds per location do standard and separable 3×3 convolutions require?
    • Standard 100; separable 576
    • Standard 576; separable 100
    • Standard 576; separable 36
    • Standard 64; separable 100
  8. In the guide’s cost ratio 1/N + 1/9 for a 3×3 separable convolution, what happens as the output-channel count N becomes very large?
    • The ratio approaches 9
    • The ratio approaches zero
    • The ratio becomes independent of kernel size for every kernel
    • The ratio approaches 1/9