Submission for Segment 3: Dataset images

[gargrohin]

Summary

• Finds the top-10 most-activating real images for each of the first 10 neurons of mixed4a in InceptionV1
• Two notebooks:
  • ImageNet-V2 (10k images, local batched DataLoader)
  • ImageNet validation set (50k images, streamed from HuggingFace with batched + threaded prefetch pipeline)
• Includes activation heatmap overlays showing where each neuron fires on its top image

Test plan

• ImageNet-V2 notebook runs end-to-end and displays outputs
• ImageNet streaming notebook runs end-to-end (requires HF_TOKEN_PERSONAL env var) on validation set
• ImageNet streaming notebook runs end-to-end (requires HF_TOKEN_PERSONAL env var) on train set
• Outputs render correctly on GitHub