Using Data Wrangler Corrupt Image Impulse Noise to Make a Fraud Camera Model Robust to Image Quality Variation
An ML engineer trained an ML model on Amazon SageMaker to detect automobile accidents from dosed-circuit TV footage. The ML engineer used SageMaker Data Wrangler to create a training dataset of images of accidents and non-accidents. The model performed well during training and validation. However, the model is underperforming in production because of variations in the quality of the images from various cameras. Which solution will improve the model's accuracy in the LEAST amount of time?
Community Votes
75% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.
Community Insight
Adding impulse noise with the Data Wrangler corrupt image transform augments the existing dataset so the model learns to predict reliably on degraded images, directly targeting the quality variation that causes the production drop-off in the fastest possible time.
A model detects automobile accidents from closed-circuit TV footage and performs well in training and validation but degrades in production because image quality varies between cameras. The requirement is to improve accuracy in the least amount of time, so the fix must reuse the existing images rather than collect more.
Choosing the enhance image contrast transform with Gamma correction because it sounds like a quality fix. Gamma adjustment changes tonal brightness but does not make the model robust to the wide quality variance across cameras, whereas noise augmentation explicitly trains that robustness.
Community Discussion (6 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Why the Answer Is Correct
The model already performs well on clean data and only fails when image quality varies, so the fastest improvement comes from making the model robust to that variation using the data already collected. The Data Wrangler corrupt image transform with the impulse noise option adds noise to the training images, so the model learns to recognize accidents even in degraded, low-quality frames. This is augmentation on the existing dataset, which is why it takes the least amount of time compared with collecting more images. Lance665 and 6913a18 both cited the AWS Data Wrangler blog post explaining that corrupting images or adding noise makes a model more robust, and eesa tied the impulse-noise option to simulating low-quality camera feeds.Why the Other Options Are Wrong
Collecting more images from all the cameras (A) requires new data collection, labeling effort, and retraining, so it is explicitly the slowest path and contradicts the least-time requirement. The enhance image contrast transform with the Gamma option (C) alters tonal brightness and clarity of the images, but it does not teach the model to handle degraded or noisy inputs, so it does not address the cross-camera quality variation that is breaking production accuracy. The resize image transform with cropping (D) standardizes image dimensions, but the production problem is quality variation, not size mismatch, so cropping to a uniform size leaves the failure mode untouched.Community Comment Notes
This was a split vote, 75 for B and 25 for C. Saransundar and a4002bd argued for C on the grounds that contrast enhancement standardizes quality across cameras, and GiorgioGss linked the same Data Wrangler blog post in support of C. The majority position, from Lance665 and 6913a18, reads the question's least-time framing as the decisive constraint and points out that noise augmentation trains robustness directly, which the contrast transform does not.Official Reference
Related Analysis
Practice All MLA-C01 Questions
Access 115 questions with complete answers and detailed explanations.
View Full MLA-C01 Practice Test →