final

abstract.tex

@@ -23,7 +23,7 @@ thus combining the representation learning benefits and speed of end-to-end deep
 with a physically plausible scene model inspired by slanted plane energy-minimization approaches to
 scene flow.
 
-Building on recent advances in region-based convolutional networks (R-CNNs),
+Building on recent advances in region-based convolutional neural networks (R-CNNs),
 we integrate motion estimation with instance segmentation.
 Given two consecutive frames from a monocular RGB-D camera,
 our resulting end-to-end deep network detects objects with precise per-pixel
@@ -54,8 +54,8 @@ Objekte respektiert, und kombinieren damit die Repräsentationskraft und Geschwi
 von end-to-end Deep Networks mit einem physikalisch plausiblen Szenenmodell,
 das von slanted-plane Energieminimierungsmethoden für Szenenfluss inspiriert ist.
 
-Hierbei bauen wir auf den aktuellen Fortschritten in regionsbasierten Convolutional
-Networks (R-CNNs) auf und integrieren Bewegungsschätzung mit Instanzsegmentierung.
+Hierbei bauen wir auf den aktuellen Fortschritten bei regionsbasierten Convolutional
+Neural Networks (R-CNNs) auf und integrieren Bewegungsschätzung mit Instanzsegmentierung.
 Bei Eingabe von zwei aufeinanderfolgenden Frames aus einer monokularen RGB-D
 Kamera erkennt unser end-to-end Deep Network Objekte mit pixelgenauen Objektmasken
 und schätzt die 3D-Bewegung jedes erkannten Objekts zwischen den Frames ab.

approach.tex

@@ -116,7 +116,7 @@ additonally dropout with $p = 0.5$ after all fully-connected hidden layers.
 }
 
 \paragraph{Motion R-CNN backbone}
-Like Faster R-CNN and Mask R-CNN, we use a ResNet \cite{ResNet} variant as backbone network to compute feature maps from input imagery.
+Like Faster R-CNN and Mask R-CNN, we use a ResNet variant \cite{ResNet} as backbone network to compute feature maps from input imagery.
 
 Inspired by FlowNetS \cite{FlowNet}, we make one modification to the ResNet backbone to enable image matching,
 laying the foundation for our motion estimation. Instead of taking a single image as input to the backbone,
@@ -125,15 +125,16 @@ Additionally, we also experiment with concatenating the camera space XYZ coordin
 XYZ$_t$ and XYZ$_{t+1}$, into the input as well.
 We do not introduce a separate network for computing region proposals and use our modified backbone network
 as both RPN and for extracting the RoI features.
+
 Technically, our feature encoder network will have to learn image matching representations similar to
-that learned by the FlowNet encoder, but the output will be computed in the
-object-centric framework of a region based convolutional network head with a 3D parametrization.
+those learned by the FlowNet encoder, but the output will be computed in the
+object-centric framework of a region-based convolutional network head with a 3D parametrization.
 Thus, in contrast to the dense FlowNet decoder, the estimated dense image matching information
-from the encoder is integrated for specific objects via RoI extraction and
+from the encoder is integrated for specific objects via RoI extraction and subsequently
 processed by the RoI head for each object.
 
 \paragraph{Per-RoI motion prediction}
-We use a rigid 3D motion parametrization similar to the one used in SE3-Nets and SfM-Net \cite{SE3Nets, SfmNet}.
+We use a 3D rigid motion parametrization similar to the one used in SE3-Nets and SfM-Net \cite{SE3Nets, SfmNet}.
 For the $k$-th object proposal, we predict the rigid transformation $\{R_k, t_k\}\in \mathbf{SE}(3)$
 \footnote{$\mathbf{SE}(3)$ refers to the Special Euclidean Group representing 3D rotations
 and translations: $\{R, t|R \in \mathbf{SO}(3), t \in \mathbb{R}^3\}$}
@@ -178,7 +179,7 @@ We then extend the Mask R-CNN head by adding a small fully-connected network for
 prediction in addition to the fully-connected layers for
 refined boxes and classes and the convolutional network for the masks.
 Like for refined boxes and masks, we make one separate motion prediction for each class.
-Each instance motion is predicted as a set of nine scalar parameters,
+Each instance motion is predicted as a set of nine values,
 $\sin(\alpha)$, $\sin(\beta)$, $\sin(\gamma)$, $t_k$ and $p_k$,
 where $\sin(\alpha)$, $\sin(\beta)$ and $\sin(\gamma)$ are clipped to $[-1, 1]$.
 Here, we assume that motions between frames are relatively small
@@ -400,12 +401,12 @@ and $(c_0, c_1, f)$ are the camera intrinsics.
 For now, the depth map is always assumed to come from ground truth.
 
 Given $k$ detections with predicted motions as above, we transform all points within the bounding
-box of a detected object according to the predicted motion of the object.
+box and mask of a detected object according to the predicted motion of the object.
 
-We first define the \emph{full image} mask $M_k$ for object k,
+For this, we first define the \emph{full image} mask $M_k$ for object k,
 which can be computed from the predicted box mask $m_k$ (for the predicted class) by bilinearly resizing
 it to the width and height of the predicted bounding box and then copying the values
-of the resized mask into a full resolution mask initialized with zeros,
+of the resized mask into a full (image) resolution mask initialized with zeros,
 starting at the top-left coordinate of the predicted bounding box.
 Again we binarize masks at a threshold of $0.5$.
 

background.tex

@@ -8,7 +8,7 @@ The optical flow
 $\mathbf{w} = (u, v)^T$ from $I_t$ to $I_{t+1}$
 maps pixel coordinates in the first frame $I_t$ to pixel coordinates of the
 visually corresponding pixel in the second frame $I_{t+1}$,
-and can be interpreted as the apparent movement of brightness patterns between the two frames.
+and can be interpreted as the (apparent) movement of brightness patterns between the two frames.
 Optical flow can be regarded as two-dimensional motion estimation.
 
 Scene flow is the generalization of optical flow to three-dimensional space and additionally
@@ -64,7 +64,7 @@ learns a spatially compressed, wide (in the number of channels) representation o
 and a fully-connected prediction network on top of the encoder.
 
 The compressed representations learned by CNNs of these categories do not, however, allow
-for prediction of high-resolution output, as spatial detail is lost through sequential applications
+for prediction of high-resolution output, as spatial detail is lost through sequential application
 of pooling or strides.
 Thus, networks for dense, high-resolution, prediction introduce a convolutional decoder on top of the representation encoder,
 performing upsampling of the compressed features and resulting in a encoder-decoder pyramid.
@@ -87,7 +87,7 @@ Recently, other, similarly generic,
 encoder-decoder CNNs have been applied to optical flow prediction as well \cite{DenseNetDenseFlow}.
 
 \subsection{SfM-Net}
-Table \ref{table:sfmnet} shows the SfM-Net \cite{SfmNet} architecture we described
+Table \ref{table:sfmnet} shows the SfM-Net architecture \cite{SfmNet} we described
 in the introduction.
 Motions and full-image masks for a fixed number N$_{motions}$ of independent objects
 are predicted in addition to a depth map, and a unsupervised re-projection loss based on
@@ -237,7 +237,7 @@ In ResNet-50, the C$_5$ bottleneck has a stride of 32 with respect to the
 input image resolution. In FlowNetS \cite{FlowNet}, their bottleneck stride is 64.
 For accurately estimating motions corresponding to larger pixel displacements, a larger
 stride may be important.
-Thus, we add a additional C$_6$ block to be used in the Motion R-CNN ResNet variants
+Thus, we add an additional C$_6$ block to be used in the Motion R-CNN ResNet variants
 to increase the bottleneck stride to 64, following FlowNetS.
 
 \subsection{Region-based CNNs}
@@ -246,14 +246,14 @@ We now give an overview of region-based convolutional networks, which are curren
 most popular deep networks for object detection, and have recently also been applied to instance segmentation.
 
 \paragraph{R-CNN}
-Region-based convolutional networks (R-CNNs) \cite{RCNN} use a non-learned algorithm external to a standard encoder CNN
+The very first region-based convolutional networks (R-CNNs) \cite{RCNN} used a non-learned algorithm external to a standard encoder CNN
 for computing \emph{region proposals} in the shape of 2D bounding boxes, which represent regions that may contain an object.
 For each of the region proposals, the input image is cropped using the region bounding box and the crop is
 passed through the CNN, which performs classification of the object (or non-object, if the region shows background).
 
 \paragraph{Fast R-CNN}
 The original R-CNN involves computing one forward pass of the CNN for each of the region proposals,
-which is costly, as there is generally a large number of proposals.
+which is costly, as there generally is a large number of proposals.
 Fast R-CNN \cite{FastRCNN} significantly reduces computation by performing only a single forward pass with the whole image
 as input to the CNN (compared to the sequential input of crops in the case of R-CNN).
 Then, fixed size (H $\times$ W) feature maps are extracted from the compressed feature map of the image,
@@ -273,31 +273,31 @@ After streamlining the CNN components, Fast R-CNN is limited by the speed of the
 algorithm, which has to be run prior to the network passes and makes up a large portion of the total
 processing time.
 The Faster R-CNN object detection system \cite{FasterRCNN} unifies the generation of region proposals and subsequent box refinement and
-classification into a single deep network, leading to faster test-time processing when compared to Fast R-CNN
+classification into a single deep network, leading to faster training and test-time processing when compared to Fast R-CNN
 and again, improved accuracy.
 This unified network operates in two stages.
 In the \emph{first stage}, one forward pass is performed on the \emph{backbone} network,
 which is a deep feature encoder CNN with the original image as input.
-Next, the \emph{backbone} output features are passed into a small, fully-convolutional \emph{Region Proposal Network (RPN)} head, which
+Next, the output features from the backbone are passed into a small, fully-convolutional \emph{Region Proposal Network} network (RPN), which
 predicts objectness scores and regresses bounding boxes at each of its output positions.
-At any of the $h \times w$ output positions of the RPN head,
-$\text{N}_a$ bounding boxes with their objectness scores are predicted as offsets relative to a fixed set of $\text{N}_a$ \emph{anchors} with different
+At any of the $h \times w$ output positions of the RPN,
+$\text{N}_a$ bounding boxes with their \emph{objectness} scores are predicted as offsets relative to a fixed set of $\text{N}_a$ \emph{anchors} with different
 aspect ratios and scales. Thus, there are $\text{N}_a \times h \times w$ reference anchors in total.
 In Faster R-CNN, $\text{N}_a = 9$, with 3 scales, corresponding
-to anchor boxes of areas of $\{128^2, 256^2, 512^2\}$ pixels and 3 aspect ratios,
+to anchor boxes of areas of $\{128^2, 256^2, 512^2\}$ pixels, and 3 aspect ratios,
 $\{1:2, 1:1, 2:1\}$. For the ResNet Faster R-CNN backbone, we generally have a stride of 16
 with respect to the input image at the RPN output (Table \ref{table:maskrcnn_resnet}).
 
 For each RPN prediction at a given position, the objectness score tells us how likely it is to correspond to a detection.
 The region proposals can then be obtained as the N highest scoring RPN predictions.
 
-Then, the \emph{second stage} corresponds to the original Fast R-CNN head network, performing classification
+Then, the \emph{second stage} corresponds to the original Fast R-CNN head, performing classification
 and bounding box refinement for each of the region proposals, which are now obtained
 from the RPN instead of being pre-computed by an external algorithm.
 As in Fast R-CNN, RoI pooling is used to extract one fixed size feature map for each of the region proposals,
 and the refined bounding boxes are predicted separately for each object class.
 
-Table~\ref{table:maskrcnn_resnet} includes an overview of the Faster R-CNN ResNet network architecture
+Table~\ref{table:maskrcnn_resnet} includes an overview of the Faster R-CNN ResNet architecture
 (for Faster R-CNN, the mask head is ignored).
 
 {
@@ -330,7 +330,7 @@ ave & average pool & N$_{RoI}$ $\times$ 2048 \\
 & From ave: fully connected, N$_{cls}$ $\cdot$ 4 & N$_{RoI}$ $\times$ N$_{cls}$ $\cdot$ 4\\
 boxes & decode bounding boxes (Eq. \ref{eq:pred_bounding_box}) & N$_{RoI}$ $\times$ N$_{cls}$ $\cdot$ 4\\
 & From ave: fully connected, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls}$ \\
-classes& softmax, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls}$ \\
+classes& softmax, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls} + 1$ \\
 \midrule
 \multicolumn{3}{c}{\textbf{RoI Head: Masks}}\\
 \midrule
@@ -352,21 +352,21 @@ whereas Faster R-CNN uses RoI pooling.
 
 \paragraph{Mask R-CNN}
 Faster R-CNN and the earlier systems detect and classify objects at bounding box granularity.
-However, it can be helpful to know class and object (instance) membership of all individual pixels,
+However, it can be helpful to know class and object (instance) membership of individual pixels,
 which generally involves computing a binary image mask for each object instance specifying which pixels belong
 to that object. This problem is called \emph{instance segmentation}.
-Mask R-CNN \cite{MaskRCNN} extends the Faster R-CNN system to instance segmentation by predicting
+Mask R-CNN \cite{MaskRCNN} extends Faster R-CNN to instance segmentation by predicting
 fixed resolution instance masks within the bounding boxes of each detected object,
 which are, at test-time, bilinearly resized to fit inside the respective bounding boxes.
 For this, Mask R-CNN simply extends the Faster R-CNN head with multiple convolutions, which
 compute a pixel-precise binary mask for each instance.
-Note that the per-class masks logits are put through a sigmoid layer, and thus there is no
+Note that the per-class masks \emph{logits} (raw network outputs) are put through a sigmoid layer, and thus there is no
 comptetition between classes in the mask prediction branch.
 
 Additionally, an important technical aspect of Mask R-CNN is the replacement of RoI pooling with
 bilinear sampling for extracting the RoI features, which is much more precise.
-In the original RoI pooling from Fast R-CNN, the bins for max-pooling are not aligned with the actual pixel
-boundary of the bounding box, and thus some detail is lost.
+In the original RoI pooling adopted from Fast R-CNN, the bins for max-pooling are not aligned with the actual pixel
+boundaries of the bounding boxes, and thus some detail is lost.
 
 The basic Mask R-CNN ResNet architecture is shown in Table \ref{table:maskrcnn_resnet}.
 
@@ -408,7 +408,7 @@ F$_1$ & $\begin{bmatrix}\textrm{fully connected}, 1024\end{bmatrix}$ $\times$ 2
 & From F$_1$: fully connected, N$_{cls}$ $\cdot$ 4 & N$_{RoI}$ $\times$ N$_{cls}$ $\cdot$ 4 \\
 boxes & decode bounding boxes (Eq. \ref{eq:pred_bounding_box}) & N$_{RoI}$ $\times$ N$_{cls}$ $\cdot$ 4\\
 & From F$_1$: fully connected, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls}$ \\
-classes& softmax, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls}$ \\
+classes& softmax, N$_{cls}$ & N$_{RoI}$ $\times$ N$_{cls} + 1$ \\
 \midrule
 \multicolumn{3}{c}{\textbf{RoI Head: Masks}}\\
 \midrule
@@ -429,12 +429,12 @@ block (see Figure \ref{figure:fpn_block}).
 
 \paragraph{Feature Pyramid Networks}
 In Faster R-CNN, a single feature map is used as the source of all RoI features during RoI extraction, independent
-of the size of the bounding box of each RoI.
+of the size of the bounding box of any specific RoI.
 However, for small objects, the C$_4$ (see Table \ref{table:resnet}) features
 might have lost too much spatial information to allow properly predicting the exact bounding
 box and a high resolution mask.
 As a solution to this, the Feature Pyramid Network (FPN) \cite{FPN} enables features
-of an appropriate scale to be used for RoI extraction, depending of the size of the bounding box of an RoI.
+of an appropriate scale to be used for RoI extraction, depending on the size of the bounding box of the RoI.
 For this, a pyramid of feature maps is created on top of the ResNet \cite{ResNet}
 encoder by combining bilinearly upsampled feature maps coming from the bottleneck
 with lateral skip connections from the encoder (Figure~\ref{figure:fpn_block}).
@@ -454,17 +454,16 @@ as the RPN heads themselves correspond to different scales.
 Now, in the RPN, higher resolution feature maps can be used for regressing smaller
 bounding boxes. For example, boxes of area close to $32^2$ are predicted using P$_2$,
 which has a stride of $4$ with respect to the input image.
-Most importantly, the RoI features can now be extracted at the pyramid level P$_j$ appropriate for a
-RoI bounding box with size $h \times w$,
+Most importantly, the RoI features can now be extracted from the pyramid level P$_j$ appropriate for a
+RoI bounding box with size $h \times w$, where
 \begin{equation}
 j = 2 + j_a,
 \end{equation}
-where
 \begin{equation}
 j_a = \mathrm{clip}\left(\left[\log_2\left(\frac{\sqrt{w \cdot h}}{s_0}\right)\right], 0, 4\right)
 \label{eq:level_assignment}
 \end{equation}
-is the index (from small anchor to large anchor) of the corresponding anchor box and
+is the index (from small anchor to large anchor) of the corresponding anchor box, and
 \begin{equation}
 s_0 = 256 \cdot 0.125
 \label{eq:level_assignment}
@@ -513,12 +512,12 @@ $c$ is the output vector from a softmax layer,
 $c_{c^*} \in (0,1)$ is the output probability for class $c^*$,
 and $\text{C}$ is the number of classes.
 Note that for the object category classifier, $\text{C} = \text{N}_{cls} + 1$,
-as in $\text{N}_{cls}$, we do not count the background class.
+as $\text{N}_{cls}$ does not include the background class.
 Finally, for multi-label classification, we define the binary (sigmoid) cross-entropy loss,
 \begin{equation}
 \ell_{cls*}(y, y^*) = -y^* \cdot \log(y) - (1 - y^*) \cdot \log(1 - y),
 \end{equation}
-where $y^* \in \{0,1\}$ is a label and $y \in (0,1)$ is the output from a sigmoid layer.
+where $y^* \in \{0,1\}$ is a ground truth label and $y \in (0,1)$ is the output of a sigmoid layer.
 Note that for the mask loss that will be introduced below, $\ell_{cls*}$ is
 the sum of the $\ell_{cls*}$-losses for all 2D positions over the mask.
 
@@ -549,7 +548,7 @@ b_h^* = \log \left( \frac{h^*}{h_r} \right),
 which represents the regression target for the bounding box
 outputs of the network.
 
-Thus, for each bounding box prediction, the network predicts the box encoding $b_e$,
+Thus, for bounding box regression, the network predicts the box encoding $b_e$,
 \begin{equation}
 b_e = (b_x, b_y, b_w, b_h),
 \end{equation}
@@ -565,7 +564,7 @@ b_w = \log \left( \frac{w}{w_r} \right),
 b_h = \log \left( \frac{h}{h_r} \right).
 \end{equation*}
 
-At test time, to get from a predicted box encoding $b_e$ to the predicted bounding box $b$,
+At test time, to convert from a predicted box encoding $b_e$ to the predicted bounding box $b$,
 the definitions above can be inverted,
 \begin{equation}
 b = (x, y, w, h),
@@ -631,14 +630,14 @@ the predicted refined RoI box encoding for class $c_i^*$.
 Additionally, for any foreground RoI, let $m_i$ be the predicted $m \times m$ mask for class $c_i^*$
 and $m_i^*$ the $m \times m$ mask target with values in $\{0,1\}$, where the mask target is cropped and resized from
 the binary ground truth mask using the RPN proposal bounding box.
-In our implementation, we use nearest neighbour resizing for resizing the mask
+In our implementation, we use nearest neighbour resizing for resizing the cropped mask
 targets.
 Note that values in $m_i$ and $c_i$ are already normalized probabilities from
 sigmoid and softmax layers, respectively.
 
 Then, the ROI loss is computed as
 \begin{equation}
-L_{RoI} = L_{cls} + L_{box} + L_{mask}
+L_{RoI} = L_{cls} + L_{box} + L_{mask},
 \end{equation}
 where
 \begin{equation}
@@ -669,7 +668,7 @@ losses are only enabled for the foreground RoIs. Note that the bounding box and
 for all classes other than $c_i^*$ are not penalized.
 
 \paragraph{Inference}
-During inference, the 300 (without FPN) or 1000 (with FPN) highest scoring region proposals
+During inference, the 300 (ResNet) or 1000 (ResNet-FPN) highest scoring region proposals
 from the RPN are selected. The corresponding features are extracted from the backbone, as during training, by using the RPN bounding boxes,
 and passed through the RoI bounding box refinement and classification heads
 (but not through the mask head).

conclusion.tex

@@ -4,7 +4,7 @@ We introduced Motion R-CNN, which enables 3D object motion estimation in paralle
 to instance segmentation in the framework of region-based convolutional networks,
 given an input of two consecutive frames (and XYZ point coordinates) from a monocular camera.
 In addition to instance motions, our network estimates the 3D ego-motion of the camera.
-We combine all these estimates to yield a dense optical flow output from our
+We combine all these estimates to obtain a dense optical flow output from our
 end-to-end deep network.
 Our model is trained on the synthetic Virtual KITTI dataset, which provides
 us with bounding box, instance mask, depth, and 3D motion ground truth,
@@ -75,8 +75,8 @@ geometry for making a more reliable depth estimate, at least when the camera
 is moving. We could also extend our method to stereo input data easily by concatenating
 all of the frames into the input image.
 In case we choose the option of integrating the depth prediction directly into
-the R-CNN,
-this would however require using a different dataset for training it, as Virtual KITTI does not
+the current backbone,
+this would however require using a different dataset for training Motion R-CNN, as Virtual KITTI does not
 provide stereo images.
 If we would use a specialized depth network, we could use stereo data
 for depth prediction and still train Motion R-CNN independently on the monocular Virtual KITTI dataset,
@@ -149,8 +149,9 @@ RoI instance motion loss. Still, to use all available information from
 ground truth optical flow and obtain more accurate supervision,
 it would likely be beneficial to add a global, flow-based camera motion loss
 independent of the RoI supervision.
-To do this, one could use a re-projection loss conceptually identical to the one
-for supervising instance motions with ground truth flow. However, to adjust for the
+To do this, one could use a re-projection loss conceptually similar to the one
+for supervising instance motions with ground truth flow,
+but computed on the full image instead of for individual RoIs. However, to adjust for the
 fact that the camera motion can only be accurately supervised with flow at positions where
 no object motion accurs, this loss would have to be masked with the ground truth
 object masks. Again, we could use this flow-based loss in an unsupervised way,

experiments.tex

@@ -27,7 +27,7 @@ from different viewing angles, resulting in a total of 10 variants per sequence.
 In addition to the RGB frames, a variety of ground truth is supplied.
 For each frame, we are given a dense depth and optical flow map and the camera
 extrinsics matrix. There are two annotated object classes, cars, and vans (N$_{cls}$ = 2).
-For all cars and vans in the each frame, we are given 2D and 3D object bounding
+For all cars and vans in each frame, we are given 2D and 3D object bounding
 boxes, instance masks, 3D poses, and various other labels.
 
 This makes the Virtual KITTI dataset ideally suited for developing our joint
@@ -85,7 +85,7 @@ $\{R_k^*, t_k^*\} \in \mathbf{SE}(3)$ as
 R_k^* = \mathrm{inv}(R_{cam}^*) \cdot R_{t+1}^k \cdot \mathrm{inv}(R_t^k),
 \end{equation}
 \begin{equation}
-t_k^* = t_{t+1}^{k}  - R_k^* \cdot t_t.
+t_k^* = t_{t+1}^{k}  - R_k^* \cdot t_t^k.
 \end{equation}
 
 As for the camera, we define $o_k^* \in \{ 0, 1 \}$,
@@ -112,11 +112,11 @@ measures the mean angle of the error rotation between predicted and ground truth
 \begin{equation}
 E_{t} = \frac{1}{N}\sum_k  \left\lVert \mathrm{inv}(R_k) \cdot (t_k^* - t_k) \right\rVert_2,
 \end{equation}
-is the mean euclidean norm between predicted and ground truth translation, and
+is the mean Euclidean distance between predicted and ground truth translation, and
 \begin{equation}
 E_{p} = \frac{1}{N}\sum_k \left\lVert p_k^* - p_k \right\rVert_2
 \end{equation}
-is the mean euclidean norm between predicted and ground truth pivot.
+is the mean Euclidean distance between predicted and ground truth pivot.
 
 Moreover, we define precision and recall measures for the detection of moving objects,
 where
@@ -151,8 +151,8 @@ Our training schedule is similar to the Mask R-CNN Cityscapes schedule \cite{Mas
 We train for a total of 192K iterations on the Virtual KITTI training set.
 For this, we use a single Titan X (Pascal) GPU and a batch size of 1,
 which results in approximately one day of training for a complete run.
-As optimizer, we use stochastic gradient descent (SGD) \cite{SGD} with a
-momentum of $0.9$.
+As optimizer, we use stochastic gradient descent (SGD) \cite{SGD} with
+momentum set to $0.9$.
 As learning rate we use $0.25 \cdot 10^{-2}$ for the
 first 144K iterations and $0.25 \cdot 10^{-3}$ for all remaining iterations.
 
@@ -255,7 +255,7 @@ the average ground truth camera translation. The camera rotation angle error
 is still relatively high, compared to the small average ground truth camera rotation.
 Although both variants use the exact same network for predicting the camera motion,
 the FPN variant performs worse here, with the error in rotation angle twice as high.
-One possible explanations that should be investigated in futher work is
+One possible explanations that should be investigated in future work is
 that in the FPN variant, all blocks in the backbone are shared between the camera
 motion branch and the feature pyramid. In the variant without FPN, the C$5$ and
 C$6$ blocks are only used in the camera branch, and thus only experience weight
@@ -265,8 +265,9 @@ As a remedy, increasing the loss weighting of the camera motion loss may be
 helpful.
 
 \paragraph{Instance motion}
-The object pivots are estimated with relatively (given that the scenes are in a realistic scale)
-high accuracy in both variants, although the FPN variant is significantly more
+The object pivots are estimated with relatively
+high accuracy in both variants (given that the scenes are in a realistic scale),
+although the FPN variant is significantly more
 accurate, which we ascribe to the higher resolution features used in this variant.
 
 The predicted 3D object translations and rotations still have a relatively high

introduction.tex

@@ -113,7 +113,7 @@ manageable pieces.
   \centering
   \includegraphics[width=\textwidth]{figures/net_intro}
 \caption{
-Overview of our network based on Mask R-CNN. For each region of interest (RoI), we predict the 3D instance motion
+Overview of our network based on Mask R-CNN \cite{MaskRCNN}. For each region of interest (RoI), we predict the 3D instance motion
 in parallel to the class, bounding box and mask. Additionally, we branch off a
 small network from the bottleneck for predicting the 3D camera ego-motion.
 Novel components in addition to Mask R-CNN are shown in red.
@@ -124,8 +124,8 @@ Novel components in addition to Mask R-CNN are shown in red.
 \subsection{Related work}
 
 In the following, we will refer to systems which use deep networks for all
-optimization and do not perform time-critical side computation (e.g. numerical optimization)
-at inference time as \emph{end-to-end} deep learning systems.
+optimization and do not perform time-critical side computation
+at inference time (e.g. numerical optimization) as \emph{end-to-end} deep learning systems.
 
 \paragraph{Deep networks in optical flow estimation}
 
@@ -142,8 +142,9 @@ where semantics become very important.
 Extensions of these approaches to scene flow estimate dense flow and dense depth
 with similarly generic networks \cite{SceneFlowDataset} and similar limitations.
 
-Other works \cite{ESI, JOF, FlowLayers, MRFlow} make use of semantic segmentation to structure
-the optical flow estimation problem and introduce reasoning at the object level,
+Other works make use of semantic segmentation to structure
+the optical flow estimation problem and introduce reasoning at the object level
+\cite{ESI, JOF, FlowLayers, MRFlow},
 but still require expensive energy minimization for each
 new input, as CNNs are only used for some of the components and numerical
 optimization is central to their inference.
@@ -168,20 +169,20 @@ without the use of (deep) learning.
 
 In a more recent approach termed Instance Scene Flow \cite{InstanceSceneFlow},
 a CNN is used to compute 2D bounding boxes and instance masks for all objects in the scene, which are then combined
-with depth obtained from a non-learned stereo algorithm, to be used as pre-computed
+with depth obtained from a non-learned stereo algorithm to be used as pre-computed
 inputs to a slanted plane scene flow model based on \cite{KITTI2015}.
 Most likely due to their use of deep learning for instance segmentation and for some other components, this
-approach outperforms the previous related scene flow methods on public benchmarks.
+approach outperforms the previous related scene flow methods on relevant public benchmarks \cite{KITTI2012, KITTI2015}.
 Still, the method uses a energy-minimization formulation for the scene flow estimation itself
 and takes minutes to make a prediction.
 
 Interestingly, the slanted plane methods achieve the current state-of-the-art
 in scene flow \emph{and} optical flow estimation on the challenging KITTI benchmarks \cite{KITTI2012, KITTI2015},
-outperforming end-to-end deep networks like \cite{SceneFlowDataset, FlowNet2}.
+outperforming end-to-end deep networks like \cite{SceneFlowDataset, FlowNet, FlowNet2}.
 However, the end-to-end deep networks are significantly faster than their energy-minimization counterparts,
 generally taking a fraction of a second instead of minutes for prediction and can often be made to run in realtime.
 These concerns restrict the applicability of the current slanted plane models in practical settings,
-which often require estimations to be done in realtime (or close to realtime) and for which an end-to-end
+which often require estimations to be done in real time (or close to real time) and for which an end-to-end
 approach based on learning would be preferable.
 
 Also, by analogy, in other contexts, the move towards end-to-end deep learning has often lead