A compact table can help a writer describe benchmark results, provided its headings say what was counted. For StagingBench, the useful distinction in this example is between a recorded framing pass, a recorded framing fail, and an unknown result.
StagingBench is an independent test of AI virtual staging tools on real room photographs. The CSV in its open dataset has 120 rows covering 12 tools and 10 rooms, with one run per tool and room cell. This is a small test, and the table should travel with that limitation wherever it is pasted.
The script below reads scores.csv from the current directory and prints Markdown to standard output. It uses Python's standard library and does not create an output file.
Keep the classification explicit
For this task, a row with alignment_ok=False is unknown. That check comes first. For the remaining rows, framing_pass=True becomes a framing pass and framing_pass=False becomes a framing fail.
This is a small script for this dataset, rather than a validator for arbitrary CSV exports. If another export changes its field names or boolean spelling, check that format before reusing the code.
import csv from collections import Counter, defaultdict counts = defaultdict(Counter) with open("scores.csv", newline="", encoding="utf-8") as source: for row in csv.DictReader(source): if row["alignment_ok"] == "False": status = "unknown" else: status = "pass" if row["framing_pass"] == "True" else "fail" counts[row["tool_dir"]][status] += 1 print("| Tool | Framing pass | Framing fail | Unknown |") print("| --- | ---: | ---: | ---: |") for tool in sorted(counts): c = counts[tool] print(f"| {tool} | {c['pass']} | {c['fail']} | {c['unknown']} |") Check the actual output
These are the first three output lines, with tools sorted alphabetically:
| Tool | Framing pass | Framing fail | Unknown | | --- | ---: | ---: | ---: | | ai-homedesign | 7 | 1 | 2 | The example row contains seven framing passes, one framing fail, and two unknowns. Its counts add to the 10 rooms in the test. Alphabetical order makes the table easy to look up without arranging products by a supposed performance hierarchy.
The header and Markdown separator are part of the output. Copying the printed text into a Markdown post produces a table, while leaving the counts available for inspection as ordinary text.
Use headings that preserve the scope
"Framing pass" is more informative here than "Success." The former names the recorded outcome being counted. The latter could suggest an overall judgment covering appearance, suitability, or other qualities the script does not assess.
Likewise, "Framing fail" should not become "Bad images." The code reads fields; it does not inspect photographs or make a visual judgment. Retaining the CSV's tool identifiers also avoids introducing an unverified mapping to product names.
"Unknown" deserves its own visible column. Removing it would make the displayed row look more complete than the evidence allows. Combining it with framing failures would make a different mistake by assigning a known outcome to an unresolved measurement.
Leave overlap out of this counter
The snippet does not read visible_overlap_pct. In the supplied CSV, that column has 26 empty values, all on unknown rows. They are never converted into zero overlap or used to assign a pass or fail. A separate overlap calculation should exclude empty values from numeric comparisons and report them explicitly.
Give the table a caption
A suitable caption would identify these as counts of recorded framing outcomes, explain that unknown means alignment_ok=False, and state that the test used 10 rooms with one run per cell.
Link the StagingBench methodology beside that caption. The table is convenient publishing material, but it cannot establish repeatability or describe the appearance of the staged photographs. Keeping the labels and the small test scope attached makes it useful without asking the counts to support more than they measure.

