ORBs Ratio vs Hamming Distance Comparison

By abhishek.verma75000 · April 5, 2026

Images with zero Hamming DCT values show a significantly higher average ORBs ratio (0.9383) compared to non-zero Hamming values (0.2608). This highlights…

The total combined computation time (descriptor generation + matcher computation) across all entries is 16,437,500,000 ns — that's about 16.44 seconds in total. On average, each entry takes approximately 16,536,720 ns, or 16.54 milliseconds, to process.

When hamming dct post resize and gray equals zero, the average orbs ratio is 0.9383. When it is non-zero, the average orbs ratio drops significantly to 0.2608. A bar chart has been generated to visually compare these two groups.

For entries where orbs ratio is above 0.8, there are 374 records. The average number of matches is 406.71, and the average hamming dct post resize and gray is 0.15. A grouped bar chart was also generated comparing these averages against entries with orbs ratio ≤ 0.8, making it easy to see how high-ratio entries differ.

There are 23 entries in the dataset with zero number good matches. These entries show a fairly consistent Hamming DCT distribution (mean of 29.8, median of 30.0, ranging from 24 to 38), suggesting moderate image dissimilarity. Their average descriptor generation time is 2.7ms and matcher computation time is 8.2ms. A histogram visualization has been generated showing the Hamming score distribution for these zero-match entries.

Out of 994 total entries, 252 (25.4%) achieve a perfect orbs ratio of 1.0. The bar chart compares these perfect entries against the remaining 742 entries across key metrics, revealing clear distinctions between the two groups.

The range of number of matches that yields the highest average orbs ratio is (168.2, 199.4], with a perfect average orbs ratio of 1.0000. This finding is based on 2 samples in that range. A bar chart has been generated showing the average orbs ratio across all 10 match-count ranges, making it easy to compare performance visually.

The scatter plot with rolling average shows how the ratio of good matches to total matches varies across the dataset index. On average, about 53% of matches are 'good' (mean=0.5301), with a median of 0.5312, indicating a fairly balanced but variable distribution across the dataset.

The orbs ratio values are spread across the full range from 0.0 to 1.0, with a mean of 0.5287 and a median of 0.5284 — meaning the distribution is nearly symmetric around the midpoint. The standard deviation of 0.3947 indicates high variability across players. Two visualizations were generated: a histogram showing the detailed shape of the distribution across 50 bins, and a bar chart breaking down counts by performance tier (Very Low, Low, Medium, High, and Excellent).

Of the 468 entries with a hamming distance below 5, an overwhelming 465 of them (99.36%) also have an ORBs ratio above 0.5. This near-perfect overlap suggests a very strong relationship between low hamming distance and high ORBs ratio in this dataset.

The analysis grouped hamming dct post resize and gray values into buckets (ranges of 5) and calculated the average orbs ratio for each. A bar chart and data table have been generated showing how the average orbs ratio varies across these buckets.

There is a very strong negative relationship between hamming dct post resize and gray and orbs ratio (correlation = -0.928). This means that as the Hamming DCT distance increases, the ORBs ratio drops significantly — lower Hamming distances indicate more similar images, which produce higher ORBs ratios. A Hamming DCT distance of 10 or less is a reliable threshold for a good match: images with Hamming ≤ 10 average an ORBs ratio of 0.814, while those above 10 average just 0.127. Two visualizations were generated — a scatter plot showing the overall relationship with threshold lines, and a box plot breaking down ORBs ratio distributions across Hamming distance ranges.

The analysis identified 70 entries that have a high number of matches (427 or more) but a low ORBs ratio (0.0643 or below), which indicates poor matching quality. A scatter plot was generated highlighting these problematic entries in red against normal entries in light blue, with threshold lines marking the Q75 match count and Q25 ratio boundaries.

The analysis examined the orbs ratio column for sudden spikes or drops that might indicate anomalous image comparisons. Three data tables were produced summarizing the findings from the dataset.