ML Interview Questions

Explain how to use pretrained models like ResNet or VGG for new computer vision tasks.

Discuss the key differences, including techniques and challenges, between 2D and 3D computer vision tasks. How do these differences impact the choice of algorithms and the complexity of real-world applications?

Explain different approaches to object detection including R-CNN, YOLO, and SSD.

Explain different data augmentation techniques and their benefits.

Describe applications of GANs in computer vision including image generation and style transfer.

Describe the evolution of object detection techniques from R-CNN to YOLO, focusing on the improvements each method introduced. Discuss the impact these advances have had on both accuracy and speed in practical applications.

Explain the evolution of object detection architectures in computer vision. Compare and contrast two-stage detectors like the R-CNN family with one-stage detectors such as YOLO and SSD. Assess their architectures, training methodologies, performance metrics like mAP and inference speed, and practical trade-offs. Additionally, discuss the application of transformers in modern object detection approaches.

Discuss the various types of image segmentation techniques such as semantic, instance, and panoptic segmentation. How do these differ in their approach and application? Compare and contrast key architectures like U-Net, Mask R-CNN, and Panoptic FPN in terms of their effectiveness, complexity, and real-world deployment.

Describe the evolution of CNN architectures for image classification from AlexNet to modern models. What key innovations improved their performance over time?

Discuss modern approaches to implementing Optical Character Recognition (OCR) using deep learning models. How do these models address challenges such as varying fonts, languages, and image distortions?