One correct option. Pick, then check.
Marr defined vision as:
Which is NOT one of Jitendra Malik's Three Rs?
Which kernel is NOT used for smoothing?
Why does JPEG use DCT instead of DFT?
Otsu's threshold maximises:
Inverse warping is preferred over forward warping because:
Why not initialise neural network weights to zero?
Kaiming (He) initialisation uses variance:
Bagging reduces ____; boosting reduces ____.
Precision is preferred over recall when:
ROC vs PR curve — which is preferred for imbalanced data with rare positives?
Conv-layer parameter count is:
Two stacked 3×3 convs at C channels have ____ params and produce a ____ receptive field.
ResNet's residual gradient ∂y/∂x=:
Depthwise-separable conv is approximately ____× cheaper than standard conv at K=3,Cout=256.
BatchNorm at inference uses:
CNNs are translation ____ but NOT ____ equivariant.
Faster R-CNN replaces Selective Search with which component?
YOLO v1 default output tensor for PASCAL VOC is:
λ_coord and λ_noobj in YOLO loss are typically:
Mask R-CNN's mask head produces:
Dilated/atrous convolutions enlarge the receptive field by:
OpenPose's output channels for K keypoints and L limb types is:
SMPL's pose parameter θ has dimensionality:
PointNet achieves permutation invariance via:
Total learnable parameters per 3DGS Gaussian (degree-3 SH):
Σ = R·S·Sᵀ·Rᵀ in 3DGS guarantees:
Why is QKᵀ divided by √dₖ?
ViT-B/16 on 224² images yields how many tokens entering the encoder (including [CLS])?
Which SSL method does NOT use negatives?
MAE's mask ratio is:
RoPE's key property is:
In PaliGemma, which component is randomly initialised?
I3D's 'inflation' trick: