Instance segmentation is a fundamental yet challenging task in computer vision, which requires an algorithm to predict a per-pixel mask with a category label for each instance of interest in an image. Despite a few works being proposed recently, the dominant framework for instance segmentation is still the two-stage method Mask R-CNN, which casts instance segmentation into a two-stage detection-and-segmentation task. Mask R-CNN first employs an object detector Faster R-CNN to predict a bounding-box for each instance. Then for each instance, regions-of-interest (ROIs) are cropped from the networks’ feature maps using the ROIAlign operation. To predict the final masks for each instance, a compact fully convolutional network (FCN) (i.e., mask head) is applied to these ROIs to perform foreground/background segmentation. However, this ROI-based method may have the following drawbacks. 1) Since ROIs are often axis-aligned bounding-boxes, for objects with irregular shapes, they may contain an excessive amount of irrelevant image content including background and other instances. This issue may be mitigated by using rotated ROIs, but with the price of a more complex pipeline. 2) In order to distinguish between the foreground instance and the background stuff or instance(s), the mask head requires a relatively larger receptive field to encode sufficiently large context information. As a result, a stack of convolutions is needed in the mask head (e.g., four convolutions with
channels in Mask R-CNN). It considerably increases computational complexity of the mask head, resulting that the inference time significantly varies in the number of instances. 3) ROIs are typically of different sizes. In order to use effective batched computation in modern deep learning frameworks[1, 23], a resizing operation is often required to resize the cropped regions into patches of the same size. For instance, Mask R-CNN resizes all the cropped regions to (upsampled to using a deconvolution), which restricts the output resolution of instance segmentation, as large instances would require higher resolutions to retain details at the boundary.
. FCNs also have shown excellent performance on many other per-pixel prediction tasks ranging from low-level image processing such as denoising, super-resolution; to mid-level tasks such as optical flow estimation and contour detection; and high-level tasks including recent single-shot object detection, monocular depth estimation  and counting . However, almost all the instance segmentation methods based on FCNs111By FCNs, we mean the vanilla FCNs in  that only involve convolutions and pooling. lag behind state-of-the-art ROI-based methods. Why do the versatile FCNs perform unsatisfactorily on instance segmentation? We observe that the major difficulty of applying FCNs to instance segmentation is that the similar image appearance may require different predictions but FCNs struggle at achieving this. For example, if two persons A and B with the similar appearance are in an input image, when predicting the instance mask of A, the FCN needs to predict B as background w.r.t. A, which can be difficult as they look similar in appearance. Therefore, the ROI operation is used to crop the person of interest, e.g., A; and filter out B. Essentially, instance segmentation needs two types of information: 1) appearance information to categorize objects; and 2) location information to distinguish multiple objects belonging to the same category. Almost all methods rely on ROI cropping, which explicitly encodes the location information of instances. In contrast, CondInst exploits the location information by using location/instance-sensitive convolution filters as well as relative coordinates that are appended to the feature map.
Thus, we advocate a new solution that uses instance-aware FCNs for instance mask prediction. In other words, instead of using a standard ConvNet with a fixed set of convolutional filters as the mask head for predicting all instances, the network parameters are adapted according to the instance to be predicted. Inspired by dynamic filtering networks  and CondConv , for each instance, a controller sub-network (see Fig. 3) dynamically generates the mask FCN network parameters (conditioned on the center area of the instance), which is then used to predict the mask of this instance. It is expected that the network parameters can encode the characteristics of this instance, and only fires on the pixels of this instance, which thus bypasses the difficulty mentioned above. These conditional mask heads are applied to the whole feature maps, eliminating the need for ROI operations. At the first glance, the idea may not work well as instance-wise mask heads may incur a large number of network parameters provided that some images contain as many as dozens of instances. However, we show that a very compact FCN mask head with dynamically-generated filters can already outperform previous ROI-based Mask R-CNN, resulting in much reduced computational complexity per instance than that of the mask head in Mask R-CNN.
We summarize our main contributions as follow.
We attempt to solve instance segmentation from a new perspective. To this end, we propose the CondInst instance segmentation framework, which achieves improved instance segmentation performance than existing methods such as Mask R-CNN while being faster. To our knowledge, this is the first time that a new instance segmentation framework outperforms recent state-of-the-art both in accuracy and speed.
CondInst is fully convolutional and avoids the aforementioned resizing operation in many existing methods, as CondInst does not rely on ROI operations. Without having to resize feature maps leads to high-resolution instance masks with more accurate edges.
Unlike previous methods, in which the filters in its mask head are fixed for all the instances once trained, the filters in our mask head are dynamically generated and conditioned on instances. As the filters are only asked to predict the mask of only one instance, it largely eases the learning requirement and thus reduces the load of the filters. As a result, the mask head can be extremely light-weight, significantly reducing the inference time per instance. Compared with the bounding box detector FCOS, CondInst needs only 10% more computational time, even processing the maximum number of instances per image (i.e., instances).
Moreover, CondInst can be immediately applied to panoptic segmentation due to its flexible design. We believe that with minimal re-design effort, the proposed CondInst can be used to solve all instance-level recognition tasks that were previously solved with an ROI-based pipeline.
1.1 Related Work
Here we review some work that is most relevant to ours.
Conditional Convolutions. Unlike traditional convolutional layers, which have fixed filters once trained, the filters of conditional convolutions are conditioned on the input and are dynamically generated by another network (i.e., a controller). This idea has been explored previously in dynamic filter networks  and CondConv  mainly for the purpose of increasing the capacity of a classification network. In this work, we extend this idea to solve the significantly more challenging task of instance segmentation.
Instance Segmentation. To date, the dominant framework for instance segmentation is still Mask R-CNN. Mask R-CNN first employs an object detector to detect the bounding-boxes of instances (e.g., ROIs). With these bounding-boxes, an ROI operation is used to crop the features of the instance from the feature maps. Finally, a compact FCN head is used to obtain the desired instance masks. Many works [5, 21, 15] with top performance are built on Mask R-CNN. Moreover, some works have explored to apply FCNs to instance segmentation. InstanceFCN  may be the first instance segmentation method that is fully convolutional. InstanceFCN proposes to predict position-sensitive score maps with vanilla FCNs. Afterwards, these score maps are assembled to obtain the desired instance masks. Note that InstanceFCN does not work well with overlapping instances. Others [25, 26, 10] attempt to first perform segmentation and the desired instance masks are formed by assembling the pixels of the same instance. To our knowledge, thus far none of these methods can outperform Mask R-CNN both in accuracy and speed on the public COCO benchmark dataset.
The recent YOLACT  and BlendMask  may be viewed as a reformulation of Mask RCNN, which decouple ROI detection and feature maps used for mask prediction. Wang et al. developed a simple FCN based instance segmentation method, showing competitive performance .
. The idea shares some similarity with CondInst in that information about an instance is encoded in the coefficients generated by FiLM. Since only the batch normalization coefficients are dynamically generated, AdaptIS needs a large mask head to achieve good performance. In contrast, CondInst directly encodes them into conv. filters of the mask head, thus having much stronger capacity. As a result, even with a very compact mask head, we believe that CondInst can achieve instance segmentation accuracy that would not be possible for AdaptIS to attain.
2 Instance Segmentation with CondInst
2.1 Overall Architecture
Given an input image , the goal of instance segmentation is to predict the pixel-level mask and the category of each instance of interest in the image. The ground-truth of instance segmentation are defined as , where is the mask for the -th instance and is the category. is on MS-COCO . Unlike semantic segmentation, which only requires to predict one mask for an input image, instance segmentation needs to predict a variable number of masks, depending on the number of instances in the image. This poses a challenge when applying traditional FCNs  to instance segmentation. In this work, our core idea is that for an image with instances, different mask heads will be dynamically generated, and each mask head will contain the characteristics of its target instance in their filters. As a result, when the mask is applied to an input, it will only fire on the pixels of the instance, thus producing the mask prediction of the instance. We illustrate the process in Fig. 1.
Recall that Mask R-CNN employs an object detector to predict the bounding-boxes of the instances in the input image. The bounding-boxes are actually the way that Mask R-CNN represents instances. Similarly, CondInst employs the instance-aware filters to represent the instances. In other words, instead of encoding the instance concept into the bounding-boxes, CondInst implicitly encodes it into the parameters of the mask heads, which is a much more flexible way. For example, it can easily represent the irregular shapes that are hard to be tightly enclosed by a bounding-box. This is one of CondInst’s advantages over the previous ROI-based methods.
Similar to the way that ROI-based methods obtain bounding-boxes, the instance-aware filters can also be obtained with an object detector. In this work, we build CondInst on the popular object detector FCOS  due to its simplicity and flexibility. Also, the elimination of anchor-boxes in FCOS can also save the number of parameters and the amount of computation of CondInst. As shown in Fig. 3, following FCOS , we make use of the feature maps of feature pyramid networks (FPNs) , whose down-sampling ratios are , , , and , respectively. As shown in Fig. 3, on each feature level of the FPN, some functional layers (in the dash box) are applied to make instance-related predictions. For example, the class of the target instance and the dynamically-generated filters for the instance. In this sense, CondInst can be viewed as the same as Mask R-CNN, both of which first attend to instances in an image and then predict the pixel-level masks of the instances (i.e., instance-first).
Besides the detector, as shown in Fig. 3, there is also a mask branch, which provides the feature maps that our generated mask heads take as inputs to predict the desired instance mask. The feature maps are denoted by . The mask branch is connected to FPN level and thus its output resolution is of the input image resolution. The mask branch has four convolutions with channels before the last layer. Afterwards, in order to reduce the number of the generated parameters, the last layer of the mask branch reduces the number of channels from to (i.e., ). Surprisingly, using can already achieve superior performance and using a larger here (e.g., 16) cannot improve the performance, as shown in our experiments. Even more aggressively, using only degrades the performance by in mask AP. Moreover, as shown in Fig. 3, is combined with a map of the coordinates, which are relative coordinates from all the locations on to the location (i.e., where the filters of the mask head are generated). Then, the combination is sent to the mask head to predict the instance mask. The relative coordinates provide a strong cue for predicting the instance mask, as shown in our experiments. Moreover, a single sigmoid is used as the final output of the mask head, and thus the mask prediction is class-agnostic. The class of the instance is predicted by the classification head in parallel with the controller, as shown in Fig. 3.
The resolution of the original mask prediction is same as the resolution of , which is of the input image resolution. In order to produce high-resolution instance masks, a bilinear upsampling is used to upsample the mask prediction by , resulting in mask prediction (if the input image size is ). We will show that the upsampling is crucial to the final instance segmentation performance of CondInst in experiments. Note that the mask’s resolution is much higher than that of Mask R-CNN (only as mentioned before).
2.2 Network Outputs and Training Targets
Similar to FCOS, each location on the FPN’s feature maps either is associated with an instance, thus being a positive sample, or is considered a negative sample. The associated instance and label for each location are determined as follows. Let us consider the feature maps and let be its down-sampling ratio. As shown in previous works [30, 28, 13], a location on the feature maps can be mapped back onto the input image as . If the mapped location falls in the center region of an instance, the location is considered to be responsible for the instance. Any locations outside the center regions are labeled as negative samples. The center region is defined as the box , where denotes the mass center of the instance, is the down-sampling ratio of and is a constant scalar being as in FCOS . As shown in Fig. 3, at a location on , CondInst has the following output heads.
Classification Head. The classification head predicts the class of the instance associated with the location. The ground-truth target is the instance’s class or (i.e., background). As in FCOS, the network predicts a
-D vectorfor the classification and each element in
corresponds to a binary classifier, whereis the number of categories.
Controller Head. The controller head, which has the same architecture as the above classification head, is used to generate the parameters of the conv. filters for the instance at the location. As mentioned before, these generated filters are used in the mask head to predict the mask of this particular instance. This is the core contribution of our work.
To predict the filters, we concatenate all the parameters of the filters (i.e., weights and biases) together as an -D vector , where is the total number of the parameters. Thus, the controller head has output channels. As mentioned before, using a very few parameters (e.g., parameters), CondInst can already achieve excellent instance segmentation performance, which not only makes the parameters can be easily generated but also results in a mask head with low computational complexity. Thus, we use a very compact FCN as the mask head, which has three convolutions, each having channels except for the last one. The generated -D vector will be reinterpreted into the weights and biases of these filters. As mentioned before, the generated filters contain information about the instance at the location, and thus the mask head with the filters will ideally only fire on the pixels of the instance, even taking as the input the whole feature maps.
Center-ness Head. The center-ness head predicts a scalar depicting the deviation from the location to the center of the target instance. The center-ness score is multiplied with the classification scores and used in NMS remove duplicated detections. We refer readers to FCOS  for the details.
Conceptually, CondInst with the above heads can already solve the instance segmentation task since CondInst needs no ROIs. However, we find that if we make use of box-based NMS, the inference time will be much reduced. Thus, we still predict bounding-boxes in CondInst. Following FCOS, CondInst also predicts a -D vector depicting the distances from the location to four sides (i.e., left, top, right and bottom) of the instance’s bounding-box. The ground-truth bounding-boxes can be easily computed from the instance’s mask annotation , and thus predicting bounding-boxes introduces no any extra annotations. We would like to highlight that the predicted bounding-boxes are only used in NMS and do not involve any ROI operations. Moreover, as shown in Table 5, the bounding-boxes prediction can be removed if the NMS using no bounding-box (e.g., mask NMS or peak NMS ) used. This is fundamentally different from previous ROI-based methods, in which the bounding-box prediction is mandatory.
2.3 Loss Function
Formally, the overall loss function of CondInst can be formulated as,
where and denote the original loss of FCOS and the loss for instance masks, respectively. being in this work is used to balance the two losses. We refer readers to FCOS for the details of . is defined as,
where is the classification label of location , which is the class of the instance associated with the location or (i.e., background) if the location is not associated with any instance. is the number of locations where . is the indicator function, being if and otherwise. is the generated filters’ parameters at location . is the combination of and a map of coordinates . As described before, is the relative coordinates from all the locations on to (i.e., the location where the filters are generated). denotes the mask head, which consists of a stack of convolutions with dynamic parameters . is the mask of the instance associated with location . is the dice loss as in , which is used to overcome the foreground-background sample imbalance. We do not employ focal loss here as it requires special initialization, which cannot be realized if the parameters are dynamically generated. Note that, in order to compute the loss between the predicted mask and the ground-truth mask , they are required to have the same size. As mentioned before, the prediction is upsampled by and thus the final prediction has half the ground-truth mask’s resolution. Thus, we downsample by to make their sizes equal. These operations are omitted in Eq. (2) for clarification.
Moreover, as shown in YOLACT , the instance segmentation task can benefit from a joint semantic segmentation task. Thus, we also conduct experiments with the joint semantic segmentation task. However, unless explicitly specified, all the experiments in the paper are without the semantic segmentation task. If used, the semantic segmentation loss is added to .
Given an input image, we forward it through the network to obtain the outputs including classification confidence , center-ness scores, box prediction and the generated parameters . We first follow the steps in FCOS to obtain the bounding-box detections. Afterwards, box-based NMS with the threshold being is used to remove duplicated detections and then the top bounding-boxes (i.e., instances) are used to compute masks. Let us assume that bounding-boxes remain after the process and thus we have groups of the generated filters. The groups of filters in turn are used in the mask head. These instance-specific mask heads are applied, in the fashion of FCNs, to the (i.e., the combination of and ) to predict the masks of the instances. Since the mask head is a very compact network (three convolutions with channels and parameters in total), the overhead of computing masks is extremely small. For example, even with detections (i.e., the maximum number of detections per image on MS-COCO), only less milliseconds in total are spent on the mask heads, which only adds computational time to the base detector FCOS. In contrast, the mask head of Mask R-CNN has four convolutions with channels, thus having more than 2.3M parameters and taking longer computational time.
We evaluate CondInst on the large-scale benchmark MS-COCO . Following the common practice [12, 30, 18], our models are trained with split train2017 (115K images) and all the ablation experiments are evaluated on split val2017 (5K images). Our main results are reported on the test-dev split (20K images).
3.1 Implementation Details
is used as our backbone network and the weights pre-trained on ImageNet are used to initialize it. For the newly added layers, we initialize them as in 
. Our models are trained with stochastic gradient descent (SGD) overV100 GPUs for 90K iterations with the initial learning rate being and a mini-batch of images. The learning rate is reduced by a factor of at iteration and , respectively. Weight decay and momentum are set as and , respectively. Following Detectron2 , the input images are resized to have their shorter sides in and their longer sides less or equal to during training. Left-right flipping data augmentation is also used during training. When testing, we do not use any data augmentation and only the scale of the shorter side being is used. The inference time in this work is measured on a single V100 GPU with image per batch.
|w/ abs. coord.||w/ rel. coord.||w/||AP||AP||AP||AP||AP||AP||AR||AR||AR|
3.2 Architectures of the Mask Head
In this section, we discuss the design choices of the mask head in CondInst. To our surprise, the performance is insensitive to the architectures of the mask head. Our baseline is the mask head of three convolutions with channels (i.e., width ). As shown in Table 1 (3rd row), it achieves in mask AP. Next, we first conduct experiments by varying the depth of the mask head. As shown in Table (a)a, apart from the mask head with depth being , all other mask heads (i.e., depth and ) attain similar performance. The mask head with depth being achieves inferior performance as in this case the mask head is actually a linear mapping, which has overly weak capacity. Moreover, as shown in Table (b)b, varying the width (i.e., the number of the channels) does not result in a remarkable performance change either as long as the width is in a reasonable range. We also note that our mask head is extremely light-weight as the filters in our mask head are dynamically generated. As shown in Table 1, our baseline mask head only takes ms per instances (the maximum number of instances on MS-COCO), which suggests that our mask head only adds small computational overhead to the base detector. Moreover, our baseline mask head only has parameters in total. In sharp contrast, the mask head of Mask R-CNN  has more than 2.3M parameters and takes computational time ( ms per instances).
3.3 Design Choices of the Mask Branch
We further investigate the impact of the mask branch. We first change , which is the number of channels of the mask branch’s output feature maps (i.e., ). As shown in Table 2, as long as is in a reasonable range (i.e., from to ), the performance keeps almost the same. is optimal and thus we use in all other experiments by default.
As mentioned before, before taken as the input of the mask heads, the mask branch’s output is concatenated with a map of relative coordinates, which provides a strong cue for the mask prediction. As shown in Table 3 (2nd row), the performance drops significantly if the relative coordinates are removed ( vs. ). The significant performance drop implies that the generated filters not only encode the appearance cues but also encode the shape of the target instance. It can also be evidenced by the experiment only using the relative coordinates. As shown in Table 3 (2rd row), only using the relative coordinates can also obtain decent performance ( in mask AP). We would like to highlight that unlike Mask R-CNN, which encodes the shape of the target instance by a bounding-box, CondInst implicitly encodes the shape into the generated filters, which can easily represent any shapes including irregular ones and thus is much more flexible. We also experiment with the absolute coordinates, but it cannot largely boost the performance as shown in Table 3 (). This suggests that the generated filters mainly carry local cues such as shapes. It is preferred to mainly rely on the local cues because we hope that CondInst is translation invariant.
3.4 How Important to Upsample Mask Predictions?
As mentioned before, the original mask prediction is upsampled and the upsampling is of great importance to the final performance. We confirm this in the experiment. As shown in Table 4, without using the upsampling (1st row in the table), in this case CondInst can produce the mask prediction with of the input image resolution, which merely achieves in mask AP because most of the details (e.g., the boundary) are lost. If the mask prediction is upsampled by factor , the performance can be significantly improved by in mask AP (from to ). In particular, the improvement on small objects is large (from to ), which suggests that the upsampling can greatly retain the details of objects. Increasing the upsampling factor to slightly worsens the performance (from to in mask AP), probably due to the relatively low-quality annotations of MS-COCO. We use factor in all other models as it has the potential to produce high-resolution instance masks.
3.5 CondInst without Bounding-box Detection
Although we still keep the bounding-box detection branch in CondInst, it is conceptually feasible to totally eliminate it if we make use of the NMS using no bounding-boxes. In this case, all the foreground samples (determined by the classification head) will be used to compute instance masks, and the duplicated masks will be removed by mask-based NMS. As shown in Table 5, with the mask-based NMS, the same overall performance can be obtained as box-based NMS ( vs. in mask AP).
3.6 Comparisons with State-of-the-art Methods
We compare CondInst against previous state-of-the-art methods on MS-COCO test-dev split. As shown in Table 6, with learning rate schedule (i.e., iterations), CondInst outperforms the original Mask R-CNN by ( vs. ). CondInst also achieves a much faster speed than the original Mask R-CNN (ms vs. ms per image on a single V100 GPU). To our knowledge, it is the first time that a new and simpler instance segmentation method, without any bells and whistles outperforms Mask R-CNN both in accuracy and speed. CondInst also obtains better performance ( vs. ) and on-par speed (ms vs ms) than the well-engineered Mask R-CNN in Detectron2 (i.e., Mask R-CNN in Table 6). Furthermore, with a longer training schedule (e.g., ) or a stronger backbone (e.g., ResNet-101), a consistent improvement is achieved as well ( vs. with ResNet-50 and vs. with ResNet-101 ), which suggests CondInst is inherently superior to Mask R-CNN. Moreover, as shown in Table 6, with the auxiliary semantic segmentation task, the performance can be boosted from to (ResNet-50) or from to (ResNet-101), without increasing the inference time. For fair comparisons, all the inference time here is measured by ourselves on the same hardware with the official codes.
We also compare CondInst with the recently-proposed instance segmentation methods. Only with half training iterations, CondInst surpasses TensorMask  by a large margin ( vs. for ResNet-50 and vs. for ResNet-101). CondInst is also faster than TensorMask (ms vs ms per image on the same GPU) with similar performance ( vs. ). Moreover, CondInst outperforms YOLACT-700  by a large margin with the same backbone ResNet-101 ( vs. and both with the auxiliary semantic segmentation task). Moreover, as shown in Fig. 2, compared with YOLACT-700 and Mask R-CNN, CondInst can preserve more details and produce higher-quality instance segmentation results.
|Mask R-CNN ||R-50-FPN||34.6||56.5||36.6||15.4||36.3||49.7|
|CondInst w/ sem.||R-50-FPN||✓||38.8||60.4||41.5||21.1||41.1||51.0|
|CondInst w/ sem.||R-101-FPN||✓||40.1||62.1||43.1||21.8||42.7||52.6|
We have proposed a new and simpler instance segmentation framework, named CondInst. Unlike previous method such as Mask R-CNN, which employs the mask head with fixed weights, CondInst conditions the mask head on instances and dynamically generates the filters of the mask head. This not only reduces the parameters and computational complexity of the mask head, but also eliminates the ROI operations, resulting in a faster and simpler instance segmentation framework. To our knowledge, CondInst is the first framework that can outperform Mask R-CNN both in accuracy and speed, without longer training schedules needed. We believe that CondInst can be a new strong alternative to Mask R-CNN for instance segmentation.
-  (2019) PyTorch: an imperative style, high-performance deep learning library. In Proc. Advances in Neural Inf. Process. Syst., pp. 8024–8035. Cited by: §1.
-  (2019) YOLACT: real-time instance segmentation. In Proc. IEEE Int. Conf. Comp. Vis., pp. 9157–9166. Cited by: Figure 2, 4th item, §1.1, §2.3, §3.6, Table 6.
-  (2016) Crowdnet: a deep convolutional network for dense crowd counting. In Proc. ACM Int. Conf. Multimedia, pp. 640–644. Cited by: §1.
-  (2020) BlendMask: top-down meets bottom-up for instance segmentation. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., Cited by: §1.1.
-  (2019) Hybrid task cascade for instance segmentation. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 4974–4983. Cited by: §1.1.
-  (2017) Deeplab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE Trans. Pattern Anal. Mach. Intell. 40 (4), pp. 834–848. Cited by: §1.
-  (2019) Tensormask: a foundation for dense object segmentation. In Proc. IEEE Int. Conf. Comp. Vis., pp. 2061–2069. Cited by: 4th item, §3.6, Table 6.
-  (2016) Instance-sensitive fully convolutional networks. In Proc. Eur. Conf. Comp. Vis., pp. 534–549. Cited by: §1.1.
-  (2009) Imagenet: a large-scale hierarchical image database. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 248–255. Cited by: §3.1.
-  (2017) Semantic instance segmentation via deep metric learning. arXiv: Comp. Res. Repository. Cited by: §1.1.
-  (2019) Rethinking imagenet pre-training. In Proc. IEEE Int. Conf. Comp. Vis., pp. 4918–4927. Cited by: Table 6.
-  (2017) Mask R-CNN. In Proc. IEEE Int. Conf. Comp. Vis., pp. 2961–2969. Cited by: Figure 2, §1, §3.2, Table 6, §3.
-  (2015) Spatial pyramid pooling in deep convolutional networks for visual recognition. IEEE Trans. Pattern Anal. Mach. Intell. 37 (9), pp. 1904–1916. Cited by: §2.2.
Deep residual learning for image recognition.
Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778. Cited by: §3.1.
-  (2019) Mask scoring r-cnn. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 6409–6418. Cited by: §1.1.
-  (2016) Dynamic filter networks. In Proc. Advances in Neural Inf. Process. Syst., pp. 667–675. Cited by: §1.1, §1.
-  (2017) Feature pyramid networks for object detection. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 2117–2125. Cited by: Figure 3, §2.1.
-  (2017) Focal loss for dense object detection. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 2980–2988. Cited by: §3.
-  (2014) Microsoft coco: common objects in context. In Proc. Eur. Conf. Comp. Vis., pp. 740–755. Cited by: §2.1, §3.
-  (2016) Learning depth from single monocular images using deep convolutional neural fields. IEEE Trans. Pattern Anal. Mach. Intell. (), pp. . Note: Cited by: §1.
-  (2018) Path aggregation network for instance segmentation. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 8759–8768. Cited by: §1.1.
-  (2015) Fully convolutional networks for semantic segmentation. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 3431–3440. Cited by: §1, §2.1, footnote 1.
-  (2016) TensorFlow: a system for large-scale machine learning. In USENIX Symp. Operating Systems Design & Implementation (OSDI), pp. 265–283. External Links: Cited by: §1.
V-net: fully convolutional neural networks for volumetric medical image segmentation. In Proc. Int. Conf. 3D Vision (3DV), pp. 565–571. Cited by: §2.3.
-  (2019) Instance segmentation by jointly optimizing spatial embeddings and clustering bandwidth. In Proc. IEEE Conf. Comp. Vis. Patt. Recogn., pp. 8837–8845. Cited by: §1.1.
-  (2017) Associative embedding: end-to-end learning for joint detection and grouping. In Proc. Advances in Neural Inf. Process. Syst., pp. 2277–2287. Cited by: §1.1.
-  (2018) Film: visual reasoning with a general conditioning layer. In Proc. AAAI Conf. Artificial Intell., Cited by: §1.1.
-  (2015) Faster r-cnn: towards real-time object detection with region proposal networks. In Proc. Advances in Neural Inf. Process. Syst., pp. 91–99. Cited by: §2.2.
-  (2019) Adaptis: adaptive instance selection network. In Proc. IEEE Int. Conf. Comp. Vis., pp. 7355–7363. Cited by: §1.1.
-  (2019) FCOS: fully convolutional one-stage object detection. In Proc. IEEE Int. Conf. Comp. Vis., pp. 9627–9636. Cited by: Figure 3, §1, §2.1, §2.2, §2.2, §3.1, §3.
-  (2019) SOLO: segmenting objects by locations. arXiv: Comp. Res. Repository. Cited by: §1.1.
-  (2019) Detectron2. Note: https://github.com/facebookresearch/detectron2 Cited by: §3.1, Table 6.
-  (2019) CondConv: conditionally parameterized convolutions for efficient inference. In Proc. Advances in Neural Inf. Process. Syst., pp. 1305–1316. Cited by: §1.1, §1.
-  (2019) Objects as points. arXiv: Comp. Res. Repository. Cited by: §2.2.