Question 1
Based on the above data, answer the given subquestions.
Deep Learning Practice End Term: 13 September 2026, Set S2 (May 2026 term)
The IIT Madras BS Deep Learning Practice (Deep Learning Practice) End Term paper sat on 13 Sept 2026, in the May 2026 term, set S2: 31 questions for 50 marks in 180 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.
- 31
- 50
- 180 min
- 18
- 4
- 9
Show answer
Correct answer: 5808
Question 2
Based on the above data, answer the given subquestions.
Show answer
Correct answer
Question 3
Based on the above data, answer the given subquestions.
Using the actual flattened feature size from previous question number 2, how many learnable parameters (weights + bias) would a fully connected layer with 10 outputs have?
Show answer
Correct answer: 58090
Question 4
Based on the above data, answer the given subquestions.
Which of the following statements about the (corrected) model are true?
Show answer
Correct answers
Question 5
Show answer
Correct answer: 24
Question 6
Show answer
Correct answer: 48
Question 7
Which layer is responsible for the decoder failing to reach the target 96×96 output, and why?
None of the layers — the decoder already produces a 96×96 output.
Show answer
Correct answer
Question 8
Show answer
Correct answer: 160
Question 9
Show answer
Correct answer: 10368
Question 10
Show answer
Correct answer: 1368
Question 11
What is the parameter compression ratio (standard ÷ depthwise-separable), rounded to 2 decimal places?
Show answer
Correct answer: 7.6 (accepted within ±0.2)
Question 12
The pointwise 1×1 layer learns combinations across the depthwise-filtered channels.
The pointwise 1×1 layer operates independently on every channel and therefore cannot mix channel information.
The parameter saving relative to a standard convolution depends on the kernel size and channel counts.
Show answer
Correct answers
The pointwise 1×1 layer learns combinations across the depthwise-filtered channels.
The parameter saving relative to a standard convolution depends on the kernel size and channel counts.
Question 13
Based on the above data, answer the given subquestions.
Which of the following statements correctly identify a real bug in the given code ?
Show answer
Correct answers
Question 14
Based on the above data, answer the given subquestions.
No effect — BatchNorm behaves identically in train and eval mode.
BatchNorm continues to use statistics from the current inference batch instead of the stored running statistics, so predictions can differ from proper evaluation-mode inference and can depend on batch composition.
The prediction will be correct but computed twice as slowly.
Show answer
Correct answer
BatchNorm continues to use statistics from the current inference batch instead of the stored running statistics, so predictions can differ from proper evaluation-mode inference and can depend on batch composition.
Question 15
Based on the above data, answer the given subquestions.
Show answer
Correct answer: 72
Question 16
A convolutional layer receives a feature map with 24 input channels and applies 40 filters, each of spatial size 5 × 5. Following the "generic level of CNN" view (a filter spans all input channels), calculate the total number of weights (excluding bias) in this layer.
Show answer
Correct answer: 24000
Question 17
A convolutional layer is applied to a 48 × 48 input using a 5 × 5 filter with zero-padding of 2 and stride 2. Using the standard convolution output-size formula, calculate the output width (one spatial dimension). The height is the same; enter one integer only.
Show answer
Correct answer: 24
Question 18
The final classification head applies a softmax over the pre-activation scores (logits). For a 3-class problem, the logits produced for a particular image are [2, 1, 0]. Compute the softmax probability assigned to the first class (round to 2 decimal places).
Show answer
Correct answer: 0.665 (accepted within ±0.005)
Question 19
Following the VGG design philosophy of replacing a single large-kernel convolution with a cascade of small 3 × 3, stride-1 convolutions, how many such 3 × 3 layers must be stacked so that the resulting stack has the same effective (one-dimensional) receptive field as a single 13 × 13 convolution
Show answer
Correct answer: 6
Question 20
Show answer
Correct answer: 7680
Question 21
The decoder of a depth-estimation network upsamples feature maps using a transposed convolution ("up-convolution") with kernel size 4, stride 2, and padding 1. If the input feature map has spatial size 24 × 24, what is the output height? The width is the same; enter one integer only.
Show answer
Correct answer: 48
Question 22
In the original U-Net used for dense prediction (and adaptable to depth estimation), each encoder stage applies two successive 3×3 convolutions with no padding (stride 1), followed by a 2×2 max-pooling with stride 2. If the input feature map to one such encoder stage has spatial size 132 × 132, what is the output height of the feature map after both convolutions and the max-pooling of that stage? The width is the same; enter one integer only.
Show answer
Correct answer: 64
Question 23
The SRGAN discriminator is a stack of 8 convolutional layers, all using 3 × 3 kernels with padding 1, and strides alternating as s = 1, 2, 1, 2, 1, 2, 1, 2 across the eight layers (the stride-1 layers preserve spatial size; the stride-2 layers downsample).
If the input image is 128 × 128, what is the output height of the feature map after the last (8th) convolutional layer, just before the dense layers? The width is the same; enter one integer only.
Show answer
Correct answer: 8
Question 24
Efficient restoration networks (e.g., LaKDNet) replace standard convolutions with depth-wise separable convolutions (a depth-wise convolution followed by a point-wise 1×1 convolution).
An input feature map of size 24 × 24 × 24 (24 channels) is processed by a depth-wise separable convolution that uses 3 × 3 depth-wise kernels and produces 48 output channels, with stride 1 and padding 1 (so the spatial size is preserved).
Calculate the total number of multiplications for the full depth-wise separable operation (depth-wise + point-wise). Give the answer in millions (e.g., if 1,234,567 write 1.23).
Show answer
Correct answer: 0.79 (accepted within ±0.01)
Question 25
Which of the following statements about the ResNet architecture and its standard training recipe are correct?
Dropout is applied after every residual block as the primary regularizer.
Batch Normalization is applied after every convolutional layer.
The identity shortcut connection lets gradients bypass the weight layers, helping mitigate vanishing gradients in very deep networks.
A global average pooling layer is used after the final convolutional stage, so the network has no large fully-connected layers other than the final classifier (FC-1000).
Show answer
Correct answers
Batch Normalization is applied after every convolutional layer.
The identity shortcut connection lets gradients bypass the weight layers, helping mitigate vanishing gradients in very deep networks.
A global average pooling layer is used after the final convolutional stage, so the network has no large fully-connected layers other than the final classifier (FC-1000).
Question 26
Which of the following statements about the GoogLeNet / Inception architecture are correct?
A naive Inception module concatenates parallel branch outputs along the channel dimension; its output depth is determined by the number of output filters/channels in the branches, not by the filter sizes.
1×1 "bottleneck" convolutions are inserted before the expensive 3×3 and 5×5 convolutions to reduce input depth and cut down the number of operations.
The final classifier relies on two large fully-connected layers (like AlexNet's FC6/FC7) which account for most of GoogLeNet's parameters.
Auxiliary classifiers are attached to intermediate layers to inject additional gradient signal and combat vanishing gradients during training.
Show answer
Correct answers
A naive Inception module concatenates parallel branch outputs along the channel dimension; its output depth is determined by the number of output filters/channels in the branches, not by the filter sizes.
1×1 "bottleneck" convolutions are inserted before the expensive 3×3 and 5×5 convolutions to reduce input depth and cut down the number of operations.
Auxiliary classifiers are attached to intermediate layers to inject additional gradient signal and combat vanishing gradients during training.
Question 27
Which of the following statements correctly describe the Region Proposal Network (RPN) in Faster R-CNN?
The RPN is fully convolutional and shares its convolutional feature maps with the downstream detection network.
The RPN relies on an external Selective Search module to generate its region proposals.
At each sliding-window location, the RPN predicts objectness scores and box refinements relative to k predefined anchor boxes.
Anchors of multiple scales and aspect ratios allow the RPN to propose regions for objects of very different sizes and shapes.
Show answer
Correct answers
The RPN is fully convolutional and shares its convolutional feature maps with the downstream detection network.
At each sliding-window location, the RPN predicts objectness scores and box refinements relative to k predefined anchor boxes.
Anchors of multiple scales and aspect ratios allow the RPN to propose regions for objects of very different sizes and shapes.
Question 28
Which of the following statements about evaluating object detectors with mAP are correct?
mAP measures only classification accuracy and is independent of how well the boxes are localized.
IoU is defined as the area of overlap divided by the area of union between the predicted box and the ground-truth box.
A prediction with IoU above the threshold and the correct class is a True Positive; one with IoU at or below the threshold (or matching no ground-truth box) is a False Positive.
Average Precision (AP) for a class is the area under its Precision–Recall curve, and mAP is the mean of AP over all classes.
Show answer
Correct answers
IoU is defined as the area of overlap divided by the area of union between the predicted box and the ground-truth box.
A prediction with IoU above the threshold and the correct class is a True Positive; one with IoU at or below the threshold (or matching no ground-truth box) is a False Positive.
Average Precision (AP) for a class is the area under its Precision–Recall curve, and mAP is the mean of AP over all classes.
Question 29
Which of the following statements correctly describe an encoder–decoder (U-Net style) architecture used for monocular depth estimation?
The encoder progressively reduces spatial resolution while increasing the number of feature channels, capturing high-level context.
Skip connections pass high-resolution features from the encoder directly to the decoder, helping recover fine spatial detail lost during downsampling.
The decoder uses transposed convolutions (or upsampling) to progressively restore the spatial resolution of the predicted depth map.
Global average pooling is applied after every decoder block to collapse spatial dimensions, since depth is a single global scalar.
Show answer
Correct answers
The encoder progressively reduces spatial resolution while increasing the number of feature channels, capturing high-level context.
Skip connections pass high-resolution features from the encoder directly to the decoder, helping recover fine spatial detail lost during downsampling.
The decoder uses transposed convolutions (or upsampling) to progressively restore the spatial resolution of the predicted depth map.
Question 30
Regarding the channel-attention module in CBAM, which of the following are correct?
It exploits the inter-channel relationship of features, learning "what" is meaningful by re-weighting feature channels.
It produces a spatial attention map of size H × W × 1 that highlights where to focus, discarding all channel information.
The spatial dimensions of the input feature map are squeezed using both average-pooling and max-pooling, producing two channel descriptors.
The pooled descriptors are passed through a shared MLP, combined, and a sigmoid produces the channel attention weights that rescale the input feature map.
Show answer
Correct answers
It exploits the inter-channel relationship of features, learning "what" is meaningful by re-weighting feature channels.
The spatial dimensions of the input feature map are squeezed using both average-pooling and max-pooling, producing two channel descriptors.
The pooled descriptors are passed through a shared MLP, combined, and a sigmoid produces the channel attention weights that rescale the input feature map.
Question 31
In the multi-scale deep network for single-image depth estimation (Eigen et al.), the architecture is split into a coarse network and a fine network. Which statement best explains their respective roles?
The coarse network uses a large receptive field (via deeper layers and fully- connected layers) to predict the global scene depth, while the fine network refines this prediction locally using input-image details.
The fine network predicts the global scene structure, and the coarse network only sharpens edges.
Both networks are architecturally identical and their outputs are simply averaged (ensembled).
The coarse network processes each pixel independently and therefore ignores global scene context.
Show answer
Correct answer
The coarse network uses a large receptive field (via deeper layers and fully- connected layers) to predict the global scene depth, while the fine network refines this prediction locally using input-image details.