Idanwo ti o sopọ mọ itọsọna · Lile Ipele

Idanwo Amunawa DETR

Test DETR object queries, Hungarian matching, no-object targets, classification and box losses, and the reason its set-based detector avoids NMS.

Jẹmọ itọsọna awọn ọnaDetr Amunawa erin
Ibeere 1 ti 8

Kini DETR yọkuro ni akawe pẹlu awọn aṣawari ibile bii Yiyara R-CNN?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini DETR yọkuro ni akawe pẹlu awọn aṣawari ibile bii Yiyara R-CNN?
    • Awọn paati ti a ṣe ni ọwọ gẹgẹbi awọn apoti oran ati idinku ti kii ṣe ti o pọju
    • Egungun ẹhin CNN patapata
    • Awọn nilo fun eyikeyi ikẹkọ data
    • Bonding apoti bi ohun o wu kika
  2. Iru algorithm wo ni DETR lo lati baramu awọn asọtẹlẹ si awọn ohun-ilẹ-otitọ lakoko ikẹkọ?
    • k-tumo si iṣupọ
    • Imudara iwọntunwọnsi
    • Hungarian (bipartite) algorithm ti o baamu
    • Iwadi ina
  3. Kini DETR's 'awọn ibeere ohun elo'?
    • Awọn ifọrọranṣẹ ti a tẹ nipasẹ olumulo
    • Awọn apoti oran ti awọn iwọn ti o wa titi
    • Awọn piksẹli ti a ṣe ayẹwo lati aworan naa
    • Eto ti o wa titi ti awọn ifibọ ti ẹkọ oluyipada yipada si wiwa
  4. Kini faaji gbogbogbo ni DETR darapọ?
    • Nẹtiwọọki loorekoore pẹlu igi ipinnu
    • Egungun ẹhin CNN pẹlu koodu oniyipada-decoder kan
    • GAN funfun kan
    • A k-sunmọ-aládùúgbò classifier
  5. How does DETR's training objective support inference without a separate NMS stage?
    • They make every possible duplicate prediction mathematically impossible
    • The model can only output one object per image
    • One-to-one matching and set losses discourage multiple predictions for the same object
    • A hidden NMS classifier supplies every training target
  6. Ewo ni ailera ti a mọ ti DETR atilẹba?
    • Ijọpọ ikẹkọ ti o lọra ati wiwa ohun-kekere alailagbara
    • Ko le ṣiṣẹ lori awọn GPU
    • Ko nilo data ti o ni aami
    • O ṣiṣẹ nikan lori awọn aworan grẹy
  7. What training label does original DETR assign to an object query that is not matched to a ground-truth object?
    • The largest visible object's class regardless of matching
    • The no-object class
    • A mandatory new ground-truth object invented for the query
    • The image's filename as a class label
  8. Which prediction errors does DETR's set loss penalize on matched object pairs?
    • Only the number of letters in the class name
    • Only the image's compression size
    • Only whether the training batch uses a GPU
    • Both the predicted class and bounding-box errors