Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele
Idanwo Action
Learn how video action-recognition models use temporal evidence, motion representations and spatiotemporal convolutions.
Ibeere 1 ti 8
Kini ni ipilẹṣẹ ṣe iyatọ idanimọ iṣe lati iyasọtọ aworan lasan?
Awotẹlẹ gbogbo ibeere8 ibeere
Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.
- Kini ni ipilẹṣẹ ṣe iyatọ idanimọ iṣe lati iyasọtọ aworan lasan?
- O ṣe idi nipa bii awọn piksẹli ṣe yipada ni gbogbo akoko, kii ṣe fireemu ẹyọkan
- O nlo awọn aworan nla
- O ṣiṣẹ nikan lori fidio dudu-ati-funfun
- Ko nilo data ikẹkọ
- Ninu nẹtiwọọki idanimọ iṣe ṣiṣan-meji Ayebaye, kini ṣiṣan keji n ṣe ilana deede?
- Awọn ọna igbi ohun
- Ṣiṣan opitika ti o nsoju išipopada
- Awọn akọle ọrọ
- GPS ipoidojuko
- Kí ni a 3D convolution fi akawe si a boṣewa 2D convolution?
- Awọn ikanni awọ
- Iwọn aworan ti o ga julọ
- Axis igba die (akoko) ki awọn asẹ fi awọn fireemu lọpọlọpọ
- A o tobi ipele iwọn
- Kini ẹtan 'afikun' ti a lo nipasẹ awoṣe I3D?
- Imọlẹ fidio ti npọ si
- Fifi awọn fẹlẹfẹlẹ diẹ sii laileto
- Fisinuirindigbindigbin faili fidio
- Gbigbe awọn iwuwo nẹtiwọọki aworan 2D sinu 3D nipa ṣiṣe ẹda wọn kọja akoko
- Kini idi ti ipilẹ data Nkan-Nkankan ṣe akiyesi fun idanimọ iṣe?
- O fi agbara mu awọn awoṣe lati ni oye itọsọna igba diẹ kuku da lori aaye ipo
- O ni awọn fidio ẹranko nikan
- Ko ni akole
- O jẹ ti awọn aworan ẹyọkan
- Ilana wo ni o ṣe iranlọwọ fun awọn awoṣe idanimọ iṣe lati kọ ẹkọ lati fidio ti ko ni aami?
- Iforukọsilẹ fireemu-nipasẹ-fireemu pẹlu ọwọ
- Awoṣe fidio iboju ti ara ẹni abojuto
- Nparẹ gbogbo data išipopada
- Lilo ohun nikan
- How many action-template classes did the initial Something-Something dataset paper describe?
- 17 classes
- 1,740 classes
- 174 classes
- 17,400 classes
- What pretraining strategy does VideoMAE use?
- Training only on single labeled frames
- Replacing video with a text transcript
- Predicting optical flow without reconstructing masked content
- Masking a high proportion of video tokens and reconstructing them from visible tokens