Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Ẹya Jibiti Networks adanwo

Test FPN’s top-down and lateral feature fusion, scale-specific levels, nearest-neighbor upsampling, and use in multiscale object detection.

Jẹmọ itọsọna awọn ọnaẸya Jibiti Networks
Ibeere 1 ti 8

Iṣoro akọkọ wo ni Nẹtiwọọki Jibiti Ẹya kan yanju ni akọkọ?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Iṣoro akọkọ wo ni Nẹtiwọọki Jibiti Ẹya kan yanju ni akọkọ?
    • Ṣiṣawari awọn nkan ti o han ni awọn iwọn ti o yatọ pupọ
    • Idinku nọmba awọn aworan ikẹkọ ti o nilo
    • Yiyọ àdáwòkọ bounding apoti lẹhin ti erin
    • Compressing si dede fun mobile imuṣiṣẹ
  2. Ti a ṣe afiwe si kikọ jibiti aworan kan (titunwọnsi ati ṣiṣiṣẹ nẹtiwọọki), FPN jẹ anfani ni pataki nitori pe:
    • Ko beere fun nẹtiwọki ẹhin rara rara
    • Tunṣe awọn ẹya ti a ṣe iṣiro tẹlẹ ninu nẹtiwọọki, fifipamọ iṣiro
    • Yiyo awọn nilo fun eyikeyi convolutions
    • Nikan ṣiṣẹ lori awọn aworan grẹy
  3. Ni FPN, kini ipa ti ọna oke-isalẹ?
    • O danu awọn ẹya ara-kekere
    • O ṣe iyipada awọn aworan si ẹya awọn ẹya ara ẹrọ fun isọdi nikan
    • O gbejade ọlọrọ ni itumọ, awọn ẹya ti o jinlẹ ṣe afẹyinti si awọn ipinnu giga
    • O ṣe ti kii-o pọju bomole
  4. Bawo ni awọn asopọ ita ṣe n ṣe idapọpọ pẹlu awọn ẹya ti oke-isalẹ ti a gbe soke?
    • Ikanni concatenated-ọlọgbọn lẹhinna jẹ idaji
    • Ilọpo eroja-ọlọgbọn lẹhin softmax kan
    • Ti yọkuro lati ṣe iṣiro maapu iyatọ
    • Fi kun eroja-ọlọgbọn lẹhin 1x1 convolution aligns awọn ikanni
  5. Kini idi ti iyipada 3x3 kan si maapu ẹya ara ẹrọ kọọkan ti a dapọ ni FPN?
    • Lati dan maapu ati ki o din aliasing lati upsampling
    • Lati dinku ipinnu aaye nipasẹ idaji
    • Lati pin aworan naa si awọn ẹka
    • Lati yi awọn ẹya pada si awọn ipoidojuko apoti-isopọ taara
  6. Kini faaji nigbamii ti o gbooro sii FPN pẹlu kikọ ẹkọ, idapọ bidirectional iwuwo?
    • AlexNet
    • BiFPN, ti a lo ninu EfficientDet
    • LeNet-5
    • Ọrọ2Vec
  7. How is a top-down FPN feature map typically resized before lateral fusion?
    • Downsampled to a single pixel
    • Rotated 90 degrees
    • Upsampled by a factor of two using nearest-neighbor interpolation
    • Converted into a text sequence
  8. What does an FPN preserve by merging shallow high-resolution features with deeper semantic features?
    • Only the original image colors
    • Only the deepest layer’s low-resolution grid
    • The exact training labels without feature extraction
    • Fine spatial detail together with stronger semantic information