Mask R-CNN
2017 · paper · AI and machine learning
Presents Mask R-CNN, which extends Faster R-CNN with a branch predicting an object mask for each detected instance. It runs at 5 fps and reports top results on COCO instance segmentation, bounding-box detection and person keypoint detection.
Checked: PDF from arxiv.org, fetched 4 Oct 2026. Page numbers are those of that PDF.
0 findings
None that passed a second check.
8 minor (typos and the like; most checked by the judge only)
Typo Appendix A: "the ResNet-FPN-50 backbone" departs from the paper's "network-depth-features" naming; the same backbone is "ResNet-50-FPN" in the Table 7 caption and elsewhere.
- p. 9 Appendix A, Implementation paragraph
the ResNet-FPN-50 backbone
- p. 9 Table 7 caption
Our method uses ResNet-50-FPN.
- p. 4 Section 3, Network Architecture
clature network-depth-features
- p. 4 Section 3, Network Architecture
nomenclature network-depth-features
Judge's note: Page 9 image shows "ResNet-FPN-50 backbone" in the Implementation paragraph and "Our method uses ResNet-50-FPN." in the Table 7 caption. The page 4 crop confirms the stated nomenclature "network-depth-features". Label inconsistency only; the model meant is unambiguous.
Second check: none (judge only)
Report a problem with this finding
Typo Section 4: "Unless noted, AP is evaluating using mask IoU." should read "AP is evaluated using mask IoU".
- p. 5 Section 4, first paragraph
Unless noted, AP is evaluating using mask IoU.
Judge's note: Page 5 image shows "Un-less noted, AP is evaluating using mask IoU." exactly as quoted. Wrong verb form; meaning unaffected.
Second check: none (judge only)
Report a problem with this finding
Typo Section 4.2: "we compare with RoIWarp proposed in MNC [10] that also adopt bilinear sampling" has a singular antecedent, so the verb should be "adopts".
- p. 6 Section 4.2, RoIAlign, second paragraph
MNC [10] that also adopt bilinear sampling
Judge's note: The crop shows the wording as quoted. The antecedent of "that" is RoIWarp (singular). Cosmetic agreement slip.
Second check: none (judge only)
Report a problem with this finding
Typo Section 5, final sentence: "we expect it be an effective framework for other instance-level tasks" is missing "to" ("expect it to be").
- p. 8 Section 5, final paragraph
we expect it be an effective framework for other instance-level tasks
Judge's note: The crop shows "we expect it / be an effective framework" across the line break with no "to" printed.
Second check: none (judge only)
Report a problem with this finding
Typo Appendix B: "With data distillation, Mask R-CNN APkp improve by 1.8 points to 69.1" should read "improves".
- p. 10 Appendix B, Keypoint Detection, second paragraph
Mask R-CNN APkp improve by 1.8 points to 69.1
Judge's note: The crop shows the wording as quoted. The number agrees with Table 9 (67.3 to 69.1), so this is a grammar slip only.
Second check: none (judge only)
Report a problem with this finding
Typo Appendix B: "This 5× increase in pre-training data improves both mask and box 1 AP." is garbled; the intended sense is "improves both mask and box AP by 1 point".
- p. 10 Appendix B, ImageNet-5k pre-training
improves both mask and box 1 AP
Judge's note: The crop shows "improves both mask and box 1 / AP" as quoted. Table 8 (ImageNet-5k row against e2e row) gives mask 37.6 to 38.6 and box 41.7 to 42.7, so the value is right and only the wording is defective.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [33] prints "P. O. Pinheiro, R. Collobert, and P. Dollar." without the accent, while [34], [35] and the title page print "Dollár".
- p. 11 Reference [33]
P. O. Pinheiro, R. Collobert, and P. Dollar.
- p. 11 Reference [34]
R. Collobert, and P. Dollár
Judge's note: The crop shows [33] with "P. Dollar" unaccented and [34] and [35] directly below with "Dollár". The page 1 image shows "Piotr Dollár" in the author line. It is an inconsistency within the paper, not an extraction loss.
Second check: none (judge only)
Report a problem with this finding
Typo References [15] and [18] print "In ECCV. 2014." with a period between venue and year, while other entries such as [28] print "In ECCV, 2014." with a comma.
- p. 11 Reference [15]
In ECCV. 2014.
- p. 11 Reference [18]
ECCV. 2014.
- p. 11 Reference [28]
In ECCV, 2014.
Judge's note: The crops show "In ECCV. 2014. 2" in [15] and "ECCV. 2014. 1, 2" in [18]. The page 11 image shows [28] "In ECCV, 2014." and [8] and [34] "In ECCV, 2016." Punctuation inconsistency only.
Second check: none (judge only)