MegaDepth
Internet photo collections paired with reconstructed depth and camera geometry.
About
MegaDepth is a public research dataset from Cornell University. Internet photo collections paired with reconstructed depth and camera geometry.
Start with a representative slice and inspect its Images, Depth, Poses signals together. This makes schema, timing, calibration, and annotation issues visible before the dataset becomes part of a larger training or evaluation pipeline.
Suggested agent workflows
Copy a prompt into your agent of choice. Each workflow starts small, respects the source terms, and makes the result inspectable in Rerun.
Audit a training slice
Fetch MegaDepth from its official source at https://www.cs.cornell.edu/projects/megadepth/. Read and preserve its source terms and citation requirements. Build a reproducible manifest for a representative slice, then load it into Rerun. Inspect the Images, Depth, Poses streams for timestamp gaps, missing samples, calibration problems, and annotation inconsistencies. Summarize the findings and save both the validation script and a Rerun recording that makes every issue easy to review.
Build a baseline pipeline
Use MegaDepth to scaffold a baseline pipeline for Monocular depth, Feature matching. Start with a small deterministic cohort and document the selection criteria. Create the dataset adapter, preprocessing steps, train and validation splits, and one lightweight baseline. Log inputs, labels, predictions, losses, and representative failures to Rerun so I can inspect the full experiment timeline. Include exact setup and reproduction commands.
Find useful failure cohorts
Analyze MegaDepth and propose a practical cohort taxonomy for Monocular depth, Feature matching. Use available metadata and the Images, Depth, Poses signals to identify hard cases, rare conditions, coverage gaps, and possible leakage between splits. Materialize the cohorts as queryable metadata, log examples from each cohort to Rerun, and produce a short report explaining which cohorts should become regression tests. Do not infer labels that are not present in the source data.