Computer Vision Interview Questions

Convolutions, detection and the geometry underneath.

25 traps, Dec 2025 to Jan 2026. Complete.

Set in interviews at OpenAI (11), Google DeepMind (7), Google (2) and Meta (2).

Each trap: the interviewer’s question, the answer most candidates give, and the mechanism that breaks it. The full answers are on Substack.

Computer Vision Engineer interview at OpenAI
The Contrastive Shortcut Trap
“We are building a 𝘡𝘦𝘳𝘰-𝘚𝘩𝘰𝘵 𝘊𝘭𝘢𝘴𝘴𝘪𝘧𝘪𝘦𝘳. We have the budget for a standard CLIP architecture. Why should we burn 25% more VRAM adding a 𝘎𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘋𝘦𝘤𝘰𝘥𝘦𝘳 (𝘊𝘰𝘊𝘢) if we don’t need to generate captions?”
Computer Vision #25
Jan 26, 2026
Senior AI Engineer interview at Google DeepMind
The Signal-to-Noise Trap
“Our competitor just trained a VLM on 6 billion image-text pairs. We only have the compute budget for 700k images. How do we beat them?”
Computer Vision #24
Jan 25, 2026
Senior AI Engineer interview at Google DeepMind
The Flamingo Architecture Trap
“We have a 70B parameter LLM. We need it to ‘see’ images. But here’s the constraint: We have zero budget to fine-tune the 70B weights, and we can’t afford to destroy the model’s existing reasoning capabilities.”
Computer Vision #23
Jan 24, 2026
final-round Computer Vision Engineer interview at OpenAI
The Interactive Segmentation Trap
“Your user clicks here. What mask does your model output?”
Computer Vision #22
Jan 23, 2026
Senior Robotics Engineer interview at NVIDIA
The Data Scaling Trap
“We need a robot to open any drawer in any user’s home. We cannot pre-train it on every possible handle shape. How do you build this?”
Computer Vision #21
Jan 22, 2026
Senior Computer Vision Engineer interview at OpenAI
The Low-Contrast Bias Trap
“Our production FaceID model has a 12% higher error rate on darker skin tones. We audited the training data and it is perfectly balanced (50/50 split). We retrained from scratch. The error persists. Why?”
Computer Vision #20
Jan 21, 2026
Senior Computer Vision Engineer interview at Google DeepMind
The Fine-Grained Invariance Trap
“We need to classify 10,000 distinct car models (Make, Model, Year) for a demographics study. How do you build the model?”
Computer Vision #19
Jan 20, 2026
Senior Computer Vision Engineer interview at Google DeepMind
The Compositionality Trap
“Our YOLO model has 99% mAP ( Mean Average Precision ) on 𝘗𝘦𝘰𝘱𝘭𝘦 and 𝘍𝘪𝘳𝘦 𝘏𝘺𝘥𝘳𝘢𝘯𝘵𝘴 individually. But in production, we saw a person sitting on a fire hydrant, and the model didn’t flag it as anomalous. It just saw two boxes. Why did we fail, and how do you fix it?”
Computer Vision #18
Jan 19, 2026
Senior AI Engineer interview at OpenAI
The Counting Hallucination Trap
“Our VLM constantly hallucinates object counts in crowded images. It says ‘8 people’ when there are only 5. We have zero budget for new data collection. How do you fix this?”
Computer Vision #17
Jan 18, 2026
Senior AI Engineer interview at OpenAI
The Contrastive Hard Negative Trap
“Our CLIP model keeps confusing Golden Retrievers with Yellow Labs. To fix it, we’re going to manually curate hard negative batches, forcing these similar breeds into the same training step. Good idea?”
Computer Vision #16
Jan 17, 2026
Senior AI Engineer interview at Meta
The Multimodal Geometry Trap
“We are building a Multimodal LLM like LLaVA. We need to feed the frozen CLIP image embeddings into our Language Model. Should we use the final [CLS] token?”
Computer Vision #15
Jan 16, 2026
AI Researcher interview at OpenAI
The Attention vs MLP Responsibility Trap
“We know 𝘚𝘦𝘭𝘧-𝘈𝘵𝘵𝘦𝘯𝘵𝘪𝘰𝘯 handles the context between tokens. So, why do we burn ~60% of our parameter budget on the 𝘗𝘰𝘴𝘪𝘵𝘪𝘰𝘯-𝘸𝘪𝘴𝘦 𝘔𝘓𝘗 𝘭𝘢𝘺𝘦𝘳𝘴? What is the MLP actually doing?”
Computer Vision #14
Jan 15, 2026
Senior Computer Vision Engineer interview at Google DeepMind
The Generalization Gap Trap
“We use heavy data augmentation (Color Jitter, 30° Rotations) during training to improve robustness. Why do we strictly disable these during validation? Doesn’t that break the rule that 𝘛𝘳𝘢𝘪𝘯 𝘢𝘯𝘥 𝘛𝘦𝘴𝘵 𝘥𝘪𝘴𝘵𝘳𝘪𝘣𝘶𝘵𝘪𝘰𝘯𝘴 𝘴𝘩𝘰𝘶𝘭𝘥 𝘮𝘢𝘵𝘤𝘩?”
Computer Vision #13
Jan 14, 2026
Senior AI Engineer interview at Google DeepMind
The Large Batch Generalization Trap
“We just scaled our infrastructure to 4x our batch size (256 to 1024) to speed up training. We followed the 𝘓𝘪𝘯𝘦𝘢𝘳 𝘚𝘤𝘢𝘭𝘪𝘯𝘨 𝘙𝘶𝘭𝘦 and multiplied our 𝘓𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘙𝘢𝘵𝘦 by 4. But our test accuracy still degraded. What fundamental property of SGD did we accidentally kill?”
Computer Vision #12
Jan 13, 2026
Senior Computer Vision Engineer interview at OpenAI
The CLIP Prompt Variance Trap
“We just deployed a CLIP model for zero-shot classification. We’re feeding in raw class names like 𝘥𝘰𝘨 or 𝘱𝘭𝘢𝘯𝘦 as text prompts. The accuracy is shaky and the variance is high. Without retraining a single parameter, 𝐡𝐨𝐰 𝐝𝐨 𝐲𝐨𝐮 𝐟𝐢𝐱 𝐭𝐡𝐞 𝐬𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐚𝐧𝐝 𝐛𝐨𝐨𝐬𝐭 𝐈𝐦𝐚𝐠𝐞𝐍𝐞𝐭 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲?”
Computer Vision #11
Jan 12, 2026
Computer Vision Engineer interview at Meta
The Early vs Slow Fusion Trap
“We’re debating between 𝘌𝘢𝘳𝘭𝘺 𝘍𝘶𝘴𝘪𝘰𝘯 and 𝘚𝘭𝘰𝘸 𝘍𝘶𝘴𝘪𝘰𝘯 for our new video understanding model. Everyone knows 𝘚𝘭𝘰𝘸 𝘍𝘶𝘴𝘪𝘰𝘯 captures motion better, but what is the specific computational consequence of maintaining that temporal dimension through multiple layers that kills our training budget?”
Computer Vision #10
Jan 11, 2026
Senior Computer Vision Engineer interview at Amazon Fulfillment Technologies & Robotics
The Tiny Object Trap
“We need to detect tiny, 3mm micro-fractures on a fast-moving assembly line. You suggested 𝐅𝐚𝐬𝐭𝐞𝐫 𝐑-𝐂𝐍𝐍 over 𝐘𝐎𝐋𝐎. Why does the 𝐑𝐞𝐠𝐢𝐨𝐧 𝐏𝐫𝐨𝐩𝐨𝐬𝐚𝐥 𝐍𝐞𝐭𝐰𝐨𝐫𝐤 (𝐑𝐏𝐍) specifically help with small objects, even though it kills our inference speed?”
Computer Vision #9
Jan 10, 2026
Senior Computer Vision Engineer interview at OpenAI
The Zero-Padding Distribution Trap
“We use Zero-Padding to maintain feature map dimensions (e.g., 32x32). But from a signal processing perspective, why is injecting zeros at the borders dangerous for your model’s statistical distribution?”
Computer Vision #8
Jan 9, 2026
Senior Computer Vision Engineer interview at OpenAI
The Receptive Field Trap
“In VGGNet, we replace a single 7x7 convolution with a stack of three 3x3 convolutions. Why?”
Computer Vision #7
Jan 8, 2026
Senior MLE Engineer interview at Google DeepMind
The Model Capacity Trap
“Our new foundation model is overfitting severely on the training set. Should we cut the hidden dimension size from 4096 to 1024 to limit its capacity?”
Computer Vision #6
Jan 7, 2026
Senior ML Engineer interview at OpenAI
The Dead ReLU Trap
“How do you fix this?”
Computer Vision #5
Jan 6, 2026
Senior Computer Vision Engineer interview at Google
The L1 vs L2 Geometry Trap
“We’re building a similarity search for a new dataset. If I arbitrarily rotate the feature space by 45 degrees, which distance metric falls apart: 𝐋1 𝐨𝐫 𝐋2? And what does that tell you about our feature engineering strategy?”
Computer Vision #4
Jan 5, 2026
Machine Learning Engineer interview at OpenAI
The Low Initial Loss Trap
“You kick off training for a Softmax classifier on CIFAR-10 (10 classes). In the very first iteration, your loss reads 0.05. Is this good news?”
Computer Vision #3
Jan 4, 2026
Senior Computer Vision Engineer interview at Google
The Redundant Data Trap
“We trained a high-capacity ResNet on 500k images, but it’s still overfitting. My Product Manager wants to spend $20k to label another 500k random images scraped from the same source. Do you approve the budget?”
Computer Vision #2
Jan 3, 2026
Senior Computer Vision Engineer interview at Tesla
The Translation Equivariance Efficiency Trap
“We all know 𝐂𝐍𝐍𝐬 are translation equivariant. But why exactly does that property make them exponentially more data-efficient than a 𝐅𝐮𝐥𝐥𝐲 𝐂𝐨𝐧𝐧𝐞𝐜𝐭𝐞𝐝 𝐧𝐞𝐭𝐰𝐨𝐫𝐤 for processing high-res images?”
Computer Vision #1
Dec 31, 2025

Get the next one