Evolving Into a Transformer: From a Training-Free Retrieval-Based Method for Anomaly Obstacle Segmentation.

As a key problem of auto-vehicle applications, the goal of Anomaly Obstacle Segmentation (AOS) is to detect some strange and unexpected obstacles (possibly are unseen previously) on the drivable area, thereby equipping the semantic perceptual model to be tolerant of unknown things. Due to its practicality, recently AOS is drawing attentions and a long line of works are proposed to tackle the obstacles with almost infinite diversity. However, these methods usually focus less on the priors of driving scenarios and involve image re-generation or the retraining of perceptual model, which lead to large computational quantity or the degradation of perceptual performance. In this paper, we propose to pay more attention to the characteristics of driving scenarios, lowering the difficulty of this tricky task. A training-free retrieval based method is thereby proposed to distinguish road obstacles from the surrounding road texture by computing the cosine similarity based on their appearance features, and significantly outperforms methods of the same category by around 20 percentage points. Besides, we find that there is a deep relation between our method and self-attention mechanism, and as a result a novel Transformer evolves from our retrieval based method, further boosting the performance.

Elastic Knowledge Distillation by Learning From Recollection.

Model performance can be further improved with the extra guidance apart from the one-hot ground truth. To achieve it, recently proposed recollection-based methods utilize the valuable information contained in the past training history and derive a "recollection" from it to provide data-driven prior to guide the training. In this article, we focus on two fundamental aspects of this method, i.e., recollection construction and recollection utilization. Specifically, to meet the various demands of models with different capacities and at different training periods, we propose to construct a set of recollections with diverse distributions from the same training history. After that, all the recollections collaborate together to provide guidance, which is adaptive to different model capacities, as well as different training periods, according to our similarity-based elastic knowledge distillation (KD) algorithm. Without any external prior to guide the training, our method achieves a significant performance gain and outperforms the methods of the same category, even as well as KD with well-trained teacher. Extensive experiments and further analysis are conducted to demonstrate the effectiveness of our method.

Memory-Efficient Class-Incremental Learning for Image Classification.

With the memory-resource-limited constraints, class-incremental learning (CIL) usually suffers from the "catastrophic forgetting" problem when updating the joint classification model on the arrival of newly added classes. To cope with the forgetting problem, many CIL methods transfer the knowledge of old classes by preserving some exemplar samples into the size-constrained memory buffer. To utilize the memory buffer more efficiently, we propose to keep more auxiliary low-fidelity exemplar samples, rather than the original real-high-fidelity exemplar samples. Such a memory-efficient exemplar preserving scheme makes the old-class knowledge transfer more effective. However, the low-fidelity exemplar samples are often distributed in a different domain away from that of the original exemplar samples, that is, a domain shift. To alleviate this problem, we propose a duplet learning scheme that seeks to construct domain-compatible feature extractors and classifiers, which greatly narrows down the above domain gap. As a result, these low-fidelity auxiliary exemplar samples have the ability to moderately replace the original exemplar samples with a lower memory cost. In addition, we present a robust classifier adaptation scheme, which further refines the biased classifier (learned with the samples containing distillation label knowledge about old classes) with the help of the samples of pure true class labels. Experimental results demonstrate the effectiveness of this work against the state-of-the-art approaches. We will release the code, baselines, and training statistics for all models to facilitate future research.

What and Where: Learn to Plug Adapters via NAS for Multidomain Learning.

As an important and challenging problem, multidomain learning (MDL) typically seeks a set of effective lightweight domain-specific adapter modules plugged into a common domain-agnostic network. Usually, existing ways of adapter plugging and structure design are handcrafted and fixed for all domains before model learning, resulting in learning inflexibility and computational intensiveness. With this motivation, we propose to learn a data-driven adapter plugging strategy with neural architecture search (NAS), which automatically determines where to plug for those adapter modules. Furthermore, we propose an NAS-adapter module for adapter structure design in an NAS-driven learning scheme, which automatically discovers effective adapter module structures for different domains. Experimental results demonstrate the effectiveness of our MDL model against existing approaches under the conditions of comparable performance.

MgSvF: Multi-Grained Slow vs. Fast Framework for Few-Shot Class-Incremental Learning.

As a challenging problem, few-shot class-incremental learning (FSCIL) continually learns a sequence of tasks, confronting the dilemma between slow forgetting of old knowledge and fast adaptation to new knowledge. In this paper, we concentrate on this '`slow vs. fast'' (SvF) dilemma to determine which knowledge components to be updated in a slow fashion or a fast fashion, and thereby balance old-knowledge preservation and new-knowledge adaptation. We propose a multi-grained SvF learning strategy to cope with the SvF dilemma from two different grains: intra-space (within the same feature space) and inter-space (between two different feature spaces). The proposed strategy designs a novel frequency-aware regularization to boost the intra-space SvF capability, and meanwhile develops a new feature space composition operation to enhance the inter-space SvF learning performance. With the multi-grained SvF learning strategy, our method outperforms the state-of-the-art approaches by a large margin.

Robust Coarse-to-Fine Registration Scheme for Mobile Laser Scanner Point Clouds Using Multiscale Eigenvalue Statistic-Based Descriptor.

To overcome the drawbacks of pairwise registration for mobile laser scanner (MLS) point clouds, such as difficulty in searching the corresponding points and inaccuracy registration matrix, a robust coarse-to-fine registration method is proposed to align different frames of MLS point clouds into a common coordinate system. The method identifies the correct corresponding point pairs from the source and target point clouds, and then calculates the transform matrix. First, the performance of a multiscale eigenvalue statistic-based descriptor with different combinations of parameters is evaluated to identify the optimal combination. Second, based on the geometric distribution of points in the neighborhood of the keypoint, a weighted covariance matrix is constructed, by which the multiscale eigenvalues are calculated as the feature description language. Third, the corresponding points between the source and target point clouds are estimated in the feature space, and the incorrect ones are eliminated via a geometric consistency constraint. Finally, the estimated corresponding point pairs are used for coarse registration. The value of coarse registration is regarded as the initial value for the iterative closest point algorithm. Subsequently, the final fine registration result is obtained. The results of the registration experiments with Autonomous Systems Lab (ASL) Datasets show that the proposed method can accurately align MLS point clouds in different frames and outperform the comparative methods.

Condition-Aware Comparison Scheme for Gait Recognition.

As an important and challenging problem, gait recognition has gained considerable attention. It suffers from confounding conditions, that is, it is sensitive to camera views, dressing types and so on. Interestingly, it is observed that, under different conditions, local body parts contribute differently to recognition performance. In this paper, we propose a condition-aware comparison scheme to measure gait pairs' similarity via a novel module named Instructor. Also, we present a geometry-guided data augmentation approach (Dresser) to enrich dressing conditions. Furthermore, to enhance the gait representation, we propose to model temporal local information from coarse to fine. Our model is evaluated on two popular benchmarks, CASIA-B and OULP. Results show that our method outperforms current state-of-the-art methods, especially in the cross-condition scenario.

Semantic Neighborhood-Aware Deep Facial Expression Recognition.

Different from many other attributes, facial expression can change in a continuous way, and therefore, a slight semantic change of input should also lead to the output fluctuation limited in a small scale. This consistency is important. However, current Facial Expression Recognition (FER) datasets may have the extreme imbalance problem, as well as the lack of data and the excessive amounts of noise, hindering this consistency and leading to a performance decreasing when testing. In this paper, we not only consider the prediction accuracy on sample points, but also take the neighborhood smoothness of them into consideration, focusing on the stability of the output with respect to slight semantic perturbations of the input. A novel method is proposed to formulate semantic perturbation and select unreliable samples during training, reducing the bad effect of them. Experiments show the effectiveness of the proposed method and state-of-the-art results are reported, getting closer to an upper limit than the state-of-the-art methods by a factor of 30% in AffectNet, the largest in-the-wild FER database by now.

Role of ATP-sensitive potassium channels and inflammatory response of basilar artery smooth muscle cells in subarachnoid hemorrhage of rabbit and immune-modulation by shikonin.

OBJECTIVE: To investigate the role of inflammatory response, oxidative damage and changes of ATP-sensitive potassium channels (sKATP) in basilar artery (BA) smooth muscle cells (SMCS) of rabbits in subarachnoid hemorrhage (SAH) model. METHODS: Time course studies on inflammatory response by real-time PCR, oxidative process and function of isolated basilar artery after SAH in New Zealand White rabbits were performed. Basilar artery smooth muscle cells (BASMCs) in each group were obtained and whole-cell patch-clamp technique was applied to record cell membrane capacitance and KATP currents. The morphologies of basal arteries were analyzed. Protective effect of shikonin were also determine by same parameters. RESULTS: Inflammatory cytokines levels were highest at 24h compare to 72h after SAH whereas the oxidative damage and cell death marker were at highest peak at 72h. Oxidative damage peak coincided with significant alterations in cell membrane capacitance, KATP currents and morphological changes in basilar arteries. Shikokin pretreatment attenuated early inflammatory response at 24h and associated oxidative damage at 72h. Finally, shikonin attenuated morphological changes in basilar arteries and dysfunction. CONCLUSION: Currents of ATP-sensitive potassium channels in basilar smooth muscle cells decreased after SAH by putative oxidative modification from immediate inflammatory response and can be protected by shikonin pretreatment.

[Effects of Lidocaine on the Rho/ROCK Signal Transduction of the Posterior Basilar Artery in Rabbit SAH].

OBJECTIVE: To observe the expression of Rho/Rho-associated kinase (ROCK) signaling pathway in the basilar artery and the effect of lidocaine on this signaling pathway after subarachnoid hemorrhage (SAH) in rabbits. METHODS: 24 New Zealand white rabbits were randomly divided into sham operation (sham) group, SAH group, and occipital cisterna lidocaine (CD) group. There were 8 rabbits in each group. Intracisternal injection of non-anticoagulant autologous arterial blood (1 mL/kg) were applied to SAH group and CD group animals to establish SAH model, sham normal saline group was injected with 37 ℃ physiological saline (1 mL/kg); after 30 min, CD group was injected with 0.3 mL 2% lidocaine cisterna, SAH group and sham group were injected with saline. After 72 h, food intake and neurologic function damage were measured. The expressions of Rho associated kinase 2 (ROCK2) and myosin light chain (MLC) and calmodulin (CaM) protein in the basilar artery were measured by Western blot. The ROCK2 and MLC and CaM gene expressions were measured by using real-time quantitative PCR. RESULTS: Compared with sham group, reduced food intake, various degrees of neurological impairment, increased ROCK2 mRNA and protein expressions in basal artery, and decreased MLC and CaM expressions were observed in SAH group and CD group (P<0.05). Compared with the SAH group, there was no statistically significant difference in diet intake and neurological damage in the CD group (P>0.05); the mRNA and protein expressions of ROCK2 in the basilar artery decreased, and the expressions of MLC and CaM increased (P<0.05). CONCLUSION: Intracisternal injection of lidocaine may inhibit the transmission of Rho/ROCK signal in the basilar artery and reduce the basilar artery smooth muscle contraction after SAH.

Baicalin Attenuates Subarachnoid Hemorrhagic Brain Injury by Modulating Blood-Brain Barrier Disruption, Inflammation, and Oxidative Damage in Mice.

In subarachnoid hemorrhagic brain injury, the early crucial events are edema formation due to inflammatory responses and blood-brain barrier disruption. Baicalin, a flavone glycoside, has antineuroinflammatory and antioxidant properties. We examined the effect of baicalin in subarachnoid hemorrhagic brain injury. Subarachnoid hemorrhage was induced through filament perforation and either baicalin or vehicle was administered 30 min prior to surgery. Brain tissues were collected 24 hours after surgery after evaluation of neurological scores. Brain tissues were processed for water content, real-time PCR, and immunoblot analyses. Baicalin improved neurological score and brain water content. Decreased levels of tight junction proteins (occludin, claudin-5, ZO-1, and collagen IV) required for blood-brain barrier function were restored to normal level by baicalin. Real-time PCR data demonstrated that baicalin attenuated increased proinflammatory cytokine (IL-1ß, IL-6, and CXCL-3) production in subarachnoid hemorrhage mice. In addition to that, baicalin attenuated microglial cell secretion of IL-1ß and IL-6 induced by lipopolysaccharide (100 ng/ml) dose dependently. Finally, baicalin attenuated induction of NOS-2 and NOX-2 in SAH mice at the mRNA and protein level. Thus, we demonstrated that baicalin inhibited microglial cell activation and reduced inflammation, oxidative damage, and brain edema.

[Neuroprotection Mechanism of Lidocaine in Rabbits with Early Brain Injury Resulted from Subarachnoid Hemorrhage].

OBJECTIVES: To determine the neuroprotection mechanism of lidocaine on early brain injury resulted from subarachnoid hemorrhage. METHODS: Eighteen New Zealand white rabbits were randomly divided into three groups: Sham group, subarachnoid hemorrhage (SAH) group and lidocaine treatment (LD) group. Operations were performed on all animals under anesthesia. Autologous nonheparinized arterial blood (1 mL/kg, body mass) was injected into cisterna magna of rabbits in the SAH and LD groups, while saline (1 mL/kg, body mass) was given to rabbits in the sham group. Thirty minutes later, intravenous injection of 0.6 mL 20 mg/mL lidocaine was given to those in the LD group, and intravenous injection of 0.6 mL saline was given to those in the Sham and SAH groups. Food intake and neurological impairments of the rabbits were assessed 72 h after the induction of SAH. The protein and mRNA experssions of Caspase-3 and cytochrome-c (Cyt-c) in hippocampus tissues were detected using real-time PCR (RT-PCR) and Western blot. RESULTS: Rabbits in the SAH and LD groups had lower food intake and higher mRNA and protein expressions of Caspase-3 and Cyt-c than those in the sham groups, which was accompanied with varying degrees of neurological impairments. No significant differences in food intake and neurological impairments were found between the SAH and LD groups (P >0.05). However, rabbits in the LD group had lower levels of mRNA and protein expressions of Caspase-3 and Cyt-c than those in the SAH group (P <0.05). CONCLUSION: The neuroprotection mechanism of lidocaine on early brain injury in rabbits with subarachnoid hemorrhage may be associated with inhibition of mitochondrial pathway and downregulated mRNA and protein expressions of Caspase-3 and Cyt-c in brain tissues.

[Comparison of Two Methods of Lidocaine Administrating for Neuroprotection in Rabbit Model of Subarachnoid Hemorrhage].

OBJECTIVES: To compare the neuroprotection effect of two methods of Lidocaine administration in rabbit model of subarachnoid hemorrhage. METHODS: Forty New Zealand white rabbits were randomly divided into sham group, subarachnoid hemorrhage group (SAH), Lidocaine intravenous injection group (L1), and Lidocaine intracisternal administration group (L2). The rabbits were given general anaesthesia, then 1.5 mL autologous nonheparinized arterial blood was injected into cisterna magna to establish SAH model, while 1.5 mL saline was used in sham group. Thirty minutes later, the rabbits in L1 and L2 group respectively received 0.3 mL 2% Lidocaine administration of intravenously and intracisternally injection. All animals were sacrificed at 72 h after SAH. The samples of basilar artery and hippocampus tissue were processed for morphometric analysis. At pre-operation and 72 h after SAH, the level of interleukin-6 (IL-6) in serum was measured. HE staining and C fos immunohistochemical staining were performed in L1 and L2 groups. Artery area and artery diameter of basal arteries, normal neuron density and C-fos positive cell in hippocampus were measured at 72 h after SAH. RESULTS: The baseline level of IL-6 was not significant different in four groups (P>0.05). The level of IL-6 at 72 h after SAH was significantly higher than that at pre-operation in SAH, L1 and L2 groups (P<0.05), while the level of IL-6 in SAH and L1 group was higher than that in L2 group (P<0.05). Compared to sham and L2 group, the cross-section area and diameter of basal artery were smaller in SAH and L1 group, while the normal neuron density of hippocampus was less (P<0.05). CONCLUSIONS: Intracisternal administration of Lidocaine could provide neuroprotection in rabbit model of subarachnoid hemorrhage.

[The effect of Xuebijing injection on immune regulation in SIRS patients].

OBJECTIVE: To investgate the effect of Xuebijing Injection on immune regulation in patients with systemic inflammatory response syndrome (SIRS). METHODS: 56 SIRS patients admitted to the ICU of Guizhou Provincial Hospital from January 2013 to December 2013 were included in this study. The patients were randomly divided into a control (C) and a treatment (T) group. Patients in C group received routine treatment; while patients in T group received additional Xuebijing Injection 50 mL Bid. The following indicators were recorded and compared between the two groups before and 4 and 7 days after treatments: CD4+, CD8+, CD4+/CD8+, monocytes CD14+/human leukocyte antigen-DR (HLA-DR), score of acute physiology and chronic health evaluation II (APACHE II), heart rate, white blood cells, body temperature, respiration rate, and MODS (7 d after treatment only). RESULTS: No differences were found between the two groups before treatments (P > 0.05). T Group had higher levels of CD4+, CD8, CD4+/CD8+ and monocytes CD14+/HLA-DR than C group at 4 and 7 d after treatments (P < 0.05). T group also had higher levels of improvement in vital indicators compared with C group (P < 0.05). No significant difference in incidence of multiple organ dysfunction syndrome (MODS) was found between the two groups (P > 0.05). CONCLUSION: Xuebijing Injection can regulate immune function of patients with SIRS.

Alterations of voltage-dependent calcium channel currents in basilar artery smooth muscle cells at early stage of subarachnoid hemorrhage in a rabbit model.

OBJECTIVE: To investigate the changes in the currents of voltage-dependent calcium channels (VDCCs) in smooth muscle cells of basilar artery in a rabbit model of subarachnoid hemorrhage (SAH). METHODS: New Zealand white rabbits were randomly divided into five groups: sham (C), normal (N), 24 hours (S1), 48 hours (S2) and 72 hours (S3) after SAH. Non-heparinized autologous arterial blood (1 ml/kg) was injected into the cisterna magna to create SAH after intravenous anesthesia, and 1 ml/kg of saline was injected into cisterna magna in the sham group. Rabbits in group N received no injections. Basilar artery in S1, S2, S3 group were isolated at 24, 48, 72 hours after SAH. Basilar artery in group C was isolated at 72 hours after physiological saline injection. Basilar artery smooth muscle cells were isolated for all groups. Whole-cell patch-clamp technique was utilized to record cell membrane capacitance and VDCCs currents. The VDCCs antagonist nifedipine was added to the bath solution to block the Ca(++) channels currents. RESULTS: There were no significant differences in the number of cells isolated, the cell size and membrane capacitance among all the five groups. VDCC currents in the S1-S3 groups had higher amplitudes than those in control and sham groups. The significant change of current amplitude was observed at 72 hours after SAH, which was higher than those of 24 and 48 hours. The VDCCs were shown to expression in human artery smooth muscle cells. CONCLUSIONS: The changes of activation characteristics and voltage-current relationship at 72 hours after SAH might be an important event which leads to a series of molecular events in the microenvironment of the basilar artery smooth muscle cells. This may be the key time point for potential therapeutic intervention against subarachnoid hemorrhage.