Recently, the rise of large-scale vision-language pretrained models like CLIP, coupled with the technology of Parameter-Efficient FineTuning (PEFT), has captured substantial attraction in video action recognition. Nevertheless, prevailing approaches tend to prioritize strong supervised performance at the expense of compromising the models' generalization capabilities during transfer. In this paper, we introduce a novel Multimodal, Multi-task CLIP adapting framework named \name to address these challenges, preserving both high supervised performance and robust transferability. Firstly, to enhance the individual modality architectures, we introduce multimodal adapters to both the visual and text branches. Specifically, we design a novel visual TED-Adapter, that performs global Temporal Enhancement and local temporal Difference modeling to improve the temporal representation capabilities of the visual encoder. Moreover, we adopt text encoder adapters to strengthen the learning of semantic label information. Secondly, we design a multi-task decoder with a rich set of supervisory signals to adeptly satisfy the need for strong supervised performance and generalization within a multimodal framework. Experimental results validate the efficacy of our approach, demonstrating exceptional performance in supervised learning while maintaining strong generalization in zero-shot scenarios.
Soft filter pruning~(SFP) has emerged as an effective pruning technique for allowing pruned filters to update and the opportunity for them to regrow to the network. However, this pruning strategy applies training and pruning in an alternative manner, which inevitably causes inconsistent representations between the reconstructed network~(R-NN) at the training and the pruned network~(P-NN) at the inference, resulting in performance degradation. In this paper, we propose to mitigate this gap by learning consistent representation for soft filter pruning, dubbed as CR-SFP. Specifically, for each training step, CR-SFP optimizes the R-NN and P-NN simultaneously with different distorted versions of the same training data, while forcing them to be consistent by minimizing their posterior distribution via the bidirectional KL-divergence loss. Meanwhile, the R-NN and P-NN share backbone parameters thus only additional classifier parameters are introduced. After training, we can export the P-NN for inference. CR-SFP is a simple yet effective training framework to improve the accuracy of P-NN without introducing any additional inference cost. It can also be combined with a variety of pruning criteria and loss functions. Extensive experiments demonstrate our CR-SFP achieves consistent improvements across various CNN architectures. Notably, on ImageNet, our CR-SFP reduces more than 41.8\% FLOPs on ResNet18 with 69.2\% top-1 accuracy, improving SFP by 2.1\% under the same training settings. The code will be publicly available on GitHub.
Semantic scene completion (SSC) aims to predict the semantic occupancy of each voxel in the entire 3D scene from limited observations, which is an emerging and critical task for autonomous driving. Recently, many studies have turned to camera-based SSC solutions due to the richer visual cues and cost-effectiveness of cameras. However, existing methods usually rely on sophisticated and heavy 3D models to directly process the lifted 3D features that are not discriminative enough for clear segmentation boundaries. In this paper, we adopt the dense-sparse-dense design and propose an end-to-end camera-based SSC framework, termed SGN, to diffuse semantics from the semantic- and occupancy-aware seed voxels to the whole scene based on geometry prior and occupancy information. By designing hybrid guidance (sparse semantic and geometry guidance) and effective voxel aggregation for spatial occupancy and geometry priors, we enhance the feature separation between different categories and expedite the convergence of semantic diffusion. Extensive experimental results on the SemanticKITTI dataset demonstrate the superiority of our SGN over existing state-of-the-art methods.
Semantic scene completion (SSC) jointly predicts the semantics and geometry of the entire 3D scene, which plays an essential role in 3D scene understanding for autonomous driving systems. SSC has achieved rapid progress with the help of semantic context in segmentation. However, how to effectively exploit the relationships between the semantic context in semantic segmentation and geometric structure in scene completion remains under exploration. In this paper, we propose to solve outdoor SSC from the perspective of representation separation and BEV fusion. Specifically, we present the network, named SSC-RS, which uses separate branches with deep supervision to explicitly disentangle the learning procedure of the semantic and geometric representations. And a BEV fusion network equipped with the proposed Adaptive Representation Fusion (ARF) module is presented to aggregate the multi-scale features effectively and efficiently. Due to the low computational burden and powerful representation ability, our model has good generality while running in real-time. Extensive experiments on SemanticKITTI demonstrate our SSC-RS achieves state-of-the-art performance.
Reliable LiDAR panoptic segmentation (LPS), including both semantic and instance segmentation, is vital for many robotic applications, such as autonomous driving. This work proposes a new LPS framework named PANet to eliminate the dependency on the offset branch and improve the performance on large objects, which are always over-segmented by clustering algorithms. Firstly, we propose a non-learning Sparse Instance Proposal (SIP) module with the ``sampling-shifting-grouping" scheme to directly group thing points into instances from the raw point cloud efficiently. More specifically, balanced point sampling is introduced to generate sparse seed points with more uniform point distribution over the distance range. And a shift module, termed bubble shifting, is proposed to shrink the seed points to the clustered centers. Then we utilize the connected component label algorithm to generate instance proposals. Furthermore, an instance aggregation module is devised to integrate potentially fragmented instances, improving the performance of the SIP module on large objects. Extensive experiments show that PANet achieves state-of-the-art performance among published works on the SemanticKITII validation and nuScenes validation for the panoptic segmentation task.
Recently, the image-wise implicit neural representation of videos, NeRV, has gained popularity for its promising results and swift speed compared to regular pixel-wise implicit representations. However, the redundant parameters within the network structure can cause a large model size when scaling up for desirable performance. The key reason of this phenomenon is the coupled formulation of NeRV, which outputs the spatial and temporal information of video frames directly from the frame index input. In this paper, we propose E-NeRV, which dramatically expedites NeRV by decomposing the image-wise implicit neural representation into separate spatial and temporal context. Under the guidance of this new formulation, our model greatly reduces the redundant model parameters, while retaining the representation ability. We experimentally find that our method can improve the performance to a large extent with fewer parameters, resulting in a more than $8\times$ faster speed on convergence. Code is available at https://github.com/kyleleey/E-NeRV.
Referring image segmentation is a typical multi-modal task, which aims at generating a binary mask for referent described in given language expressions. Prior arts adopt a bimodal solution, taking images and languages as two modalities within an encoder-fusion-decoder pipeline. However, this pipeline is sub-optimal for the target task for two reasons. First, they only fuse high-level features produced by uni-modal encoders separately, which hinders sufficient cross-modal learning. Second, the uni-modal encoders are pre-trained independently, which brings inconsistency between pre-trained uni-modal tasks and the target multi-modal task. Besides, this pipeline often ignores or makes little use of intuitively beneficial instance-level features. To relieve these problems, we propose MaIL, which is a more concise encoder-decoder pipeline with a Mask-Image-Language trimodal encoder. Specifically, MaIL unifies uni-modal feature extractors and their fusion model into a deep modality interaction encoder, facilitating sufficient feature interaction across different modalities. Meanwhile, MaIL directly avoids the second limitation since no uni-modal encoders are needed anymore. Moreover, for the first time, we propose to introduce instance masks as an additional modality, which explicitly intensifies instance-level features and promotes finer segmentation results. The proposed MaIL set a new state-of-the-art on all frequently-used referring image segmentation datasets, including RefCOCO, RefCOCO+, and G-Ref, with significant gains, 3%-10% against previous best methods. Code will be released soon.
Recently, Space-Time Memory Network (STM) based methods have achieved state-of-the-art performance in semi-supervised video object segmentation (VOS). A critical problem in this task is how to model the dependency both among different frames and inside every frame. However, most of these methods neglect the spatial relationships (inside each frame) and do not make full use of the temporal relationships (among different frames). In this paper, we propose a new transformer-based framework, termed TransVOS, introducing a vision transformer to fully exploit and model both the temporal and spatial relationships. Moreover, most STM-based approaches employ two disparate encoders to extract features of two significant inputs, i.e., reference sets (history frames with predicted masks) and query frame, respectively, increasing the models' parameters and complexity. To slim the popular two-encoder pipeline while keeping the effectiveness, we design a single two-path feature extractor to encode the above two inputs in a unified way. Extensive experiments demonstrate the superiority of our TransVOS over state-of-the-art methods on both DAVIS and YouTube-VOS datasets. Codes will be released when it is published.