MearcAll papers

arXiv 1703.06870 · cs.CV

Mask R-CNN

Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick

Claims were extracted automatically by reducer-1 on 2026-11-12. The paper's license does not allow republishing it, so long claims are shortened to a quotation. Read the full text on arXiv.

No reproduction recordedclm_4efbfd19
Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_37ede54b
However, we show that a surprisingly simple, flexible, and fast system can surpass prior state-of-the-art instance segmentation results.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_71b84ce5
To fix the misalignment, we propose a simple, quantization-free layer, called RoIAlign, that faithfully preserves exact spatial locations.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_c6315b12
Masks are shown in color, and bounding box, category, and confidences are also shown. a seemingly minor change, RoIAlign has a large impact: it improves mask accuracy by relative 10% … (excerpt)

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_2a2e12bc
Mask R-CNN surpasses the winner of the 2016 COCO keypoint competition, and at the same time runs at 5 fps.

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_761e996b
Next, we introduce the key elements of Mask R-CNN, including pixel-to-pixel alignment, which is the main missing piece of Fast/Faster R-CNN.

Mask R-CNN · page 3 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_cb1ac55b
We show by experiments that this formulation is key for good instance segmentation results.

Mask R-CNN · page 3 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_6ae587c2
To address this, we propose an RoIAlign layer that removes the harsh quantization of RoIPool, properly aligning the extracted features with the input.

Mask R-CNN · page 3 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_7417a302
RoIAlign leads to large improvements as we show in §4.2.

Mask R-CNN · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_3e6c07d7
Note that since we only compute masks on the top 100 detection boxes, Mask R-CNN adds a small overhead to its Faster R-CNN counterpart (e.g., ∼20% on typical models).

Implementation Details · page 5 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_c9cdbb1a
We perform a thorough comparison of Mask R-CNN to the state of the art along with comprehensive ablations on the COCO dataset [28].

Experiments: Instance Segmentation · page 5 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_eb9776f7
All instantiations of our model outperform baseline variants of previous state-of-the-art models.

Main Results · page 5 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_9518aeaa
Our RoIAlign layer improves AP by ∼3 points and AP75 by∼5 points.

Main Results · page 6 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_12a36433
RoIAlign improves AP by about 3 points over RoIPool, with much of the gain coming at high IoU (AP 75).

Ablation Experiments · page 6 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_3da3109a
As can be seen in Table 2c, RoIWarp performs on par with RoIPool and much worse than RoIAlign.

Ablation Experiments · page 6 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_5176e4c8
Mask R-CNN using ResNet-101- FPN outperforms the base variants of all previous state-of-the-art models (the mask output is ignored in these experiments).

Ablation Experiments · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_233457b8
As a further comparison, we trained a version of Mask R-CNN but without the mask branch, denoted by “Faster R-CNN, RoIAlign” in Table 3.

Bounding Box Detection Results · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_49d93b9d
On the other hand, it is 0.9 points box AP lower than Mask R-CNN.

Bounding Box Detection Results · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_068b0f85
This model runs at 195ms per image on an Nvidia Tesla M40 GPU (plus 15ms CPU time resizing the outputs to the original resolution), and achieves statistically the same mask … (excerpt)

Timing · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_3a9a20d7
Although Mask R-CNN is fast, we note that our design is not optimized for speed, and better speed/accuracy tradeoffs could be achieved [21],e.g., by varying image sizes and proposal numbers, … (excerpt)

Timing · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_522c626b
Adding a segment branch (for the person category) improves the AP kp to 63.1 (Table 4) on test-dev.

Mask R-CNN for Human Pose Estimation · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_27d4bd9e
Though this ResNet-50-FPN backbone has finer strides (e.g., 4 pixels on the finest level), RoIAlign still shows significant improvement over RoIPool and increases APkp by 4.4 points.

Mask R-CNN for Human Pose Estimation · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_16c5e90d
Without using the coarse training set, our method achieves 26.2 AP on test, which is over 30% relative improvement over the previous best entry (DIN [3]), and is also better … (excerpt)

Mask R-CNN for Human Pose Estimation · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_2d7e0608
Compared to the best entry using fine data only (17.4 AP), we achieve a ∼50% improvement.

Mask R-CNN for Human Pose Estimation · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_a54e4f80
We argue that within-category overlap is a core difficulty of instance segmentation.

Mask R-CNN for Human Pose Estimation · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_be5e01e5
Our method shows massive improvement on these two categories over the other best entries (relative ∼40% improvement on person from 21.8 to 30.5 and ∼20% improvement on car from 39.4 … (excerpt)

Mask R-CNN for Human Pose Estimation · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_255ecd8a
We show that using COCO pretraining is an effective strategy on this dataset.

AP on test, almost a 6 point improvement over the · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_2009dd13
COCO pre-training helps to improve results the most on these categories; however, the domain shift persists with 38.0/30.1, 57.5/40.9, and 41.2/30.9 val/test AP, respectively.

AP on test, almost a 6 point improvement over the · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_1a16157f
Thanks to its generality and flexibility, Mask R-CNN was used as the framework by the three winning teams in the COCO 2017 instance segmentation competition, which all significantly outperformed the … (excerpt)

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_b19ab567
Overall, the improvements increase mask AP 5.1 points (from 36.7 to 41.8) and box AP 7.7 points (from 39.6 to 47.3).

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_6a157067
Table 8 shows that e2e training improves mask AP by 0.6 and box AP by 1.2.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_9a89bce3
This 5× increase in pre-training data improves both mask and box 1 AP.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_166717e1
Train-time augmentation improves mask AP by 0.6 and box AP by 0.8.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_cbd8f0f6
Using the recently proposed non-local (NL) model [43], we achieve 40.3 mask AP and 45.0 box AP.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_f5354a32
This improves AP kp by about 1 point.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_e508bb6a
With data distillation, Mask R-CNN AP kp improve by 1.8 points to 69.1.

AP on test, almost a 6 point improvement over the · page 10 · Automated extraction, not reviewed · 0 reviewed reproductions

Reproduced one of these claims?

Record your result in the Mearc app on your own machine, then use Publish this result with this paper's id (1703.06870) and the claim id shown next to the claim. Mearc creates a file that holds only what you choose to publish. Add it to the public claim data by pull request; a maintainer other than you reviews it before it counts.