Offline-to-Online Knowledge Distillation for Video Instance Segmentation
2
Citation
42
Reference
10
Related Paper
Citation Trend
Abstract:
In this paper, we present offline-to-online knowledge distillation (OOKD) for video instance segmentation (VIS), which transfers a wealth of video knowledge from an offline model to an online model for consistent prediction. Unlike previous methods that have adopted either an online or offline model, our single online model takes advantage of both models by distilling offline knowledge. To transfer knowledge correctly, we propose query filtering and association (QFA), which filters irrelevant queries to exact instances. Our KD with QFA increases the robustness of feature matching by encoding object-centric features from a single frame supplemented by long-range global information. We also propose a simple data augmentation scheme for knowledge distillation in the VIS task that fairly transfers the knowledge of all classes into the online model. Extensive experiments show that our method significantly improves the performance in video instance segmentation, especially for challenging datasets, including long, dynamic sequences. Our method also achieves state-of-the-art performance on YTVIS-21, YTVIS-22, and OVIS datasets, with mAP scores of 46.1%, 43.6%, and 31.1%, respectively.According to the 2010 Malawi Demographic and Health Survey (MDHS), about 65% of households in Malawi do not have access to treated water. Distillation is one technique used for treating water. Many distillation methods are available but they are either energy intensive or contribute to environmental degradation due to their nature. However, solar energy can be used as an alternative source of energy for water distillation. There are many designs of solar distillation systems but the most-widely used one is the conventional still. Internal surfaces of the walls of the conventional solar still (CSS) are commonly painted black to avert condensation of water vapor on the walls. However, the CSS suffers from low production of distilled water and there is, therefore, a need to improve its performance. In this study, two conventional stills were designed with an identical geometry but the internal surfaces of their walls were painted white. These solar stills were tested outdoors under the same meteorological conditions at the Malawi Polytechnic (15° 42' S, 35°02' E). Distillate output was measured during experimentation. It was found that the average daily distillate outputs were 2.55 kgm-2 and 2.38 kgm-2 for the experimental still and CSS respectively. In addition, the efficiency of the experimental solar still was 6.8% more than that of the CSS. It can therefore be concluded that painting the internal surfaces of the walls of the still white improves the distillate output of the still.
Solar still
Distilled water
Cite
Citations (31)
Solar distillation can be used effectively to produce portable water using sea water, but its low efficiency has restricted its utilization. It could be still employed at hot climatic geographical locations such as Manipal effectively to produce portable drinking water. The purpose of this paper is to study the effect of different absorbing materials on distillate production under the climatic condition of Manipal so as achieve maximum distillate. Single-basin double slope solar still with an effective insolation area of 0.6m2 was used to carry out the investigation. Experimental results as expected showed that the distillate production increased with different absorbing materials as compared to distillate production without any absorbing materials. Ink and dye were used as absorbing materials and results were compared against distillation with any absorbing materials.
Solar still
Cite
Citations (6)
Reboiler
Continuous distillation
Zeotropic mixture
Batch distillation
Sequence (biology)
Component (thermodynamics)
Cite
Citations (4)
As an important separation unit, distillation column is widely applied in petrochemical and other process industry. For separating multicomponent mixtures, distillation is conducted sequentially in industry. Both individual column and distillation sequence optimization are efficient ways for saving energy consumption. Distillation sequence is usually evaluated by number of distillation subproblems, subgroups and distillation sequences. Distillation sequence has been well studied based on simple column assumption. Dividing wall column (DWC), which is an atypical distillation column for separating a multicomponent feed mixture into three output streams, as a thermally coupled distillation column, has been proposed and applied in distillation sequence. Usually sharp split is also assumed in most literatures on DWCs. A distillation sequence with DWC will give more number of feasible sequences. It is important to estimate the total available number of distillation sequences theoretically. In this work, distillation sequences with both simple column and DWC are considered. Inferential deduction method has been used to explore the number of distillation sequences for multi-component sharp splits. The three general term formulas are obtained with the assumption of sharp split. Under different assumptions, the corresponding numbers of distillation sequences are also discussed.
Continuous distillation
Petrochemical
Sequence (biology)
Batch distillation
Cite
Citations (2)
In this paper, we propose new evaluation measures for scene segmentation results, which are based on computing the difference between a region extracted from a segmentation map and the corresponding one on an ideal segmentation. The proposed measures take into account separately both under and over detected pixels. It also associates in its computation the compactness of the region under investigation.
Segmentation-based object categorization
Region growing
Cite
Citations (23)
Consistent segmentation of COVID-19 patient's CT scans across multiple time points is essential to assess disease progression and response to therapy accurately. Existing automatic and interactive segmentation models for medical images only use data from a single time point (static). However, valuable segmentation information from previous time points is often not used to aid the segmentation of a patient's follow-up scans. Also, fully automatic segmentation techniques frequently produce results that would need further editing for clinical use. In this work, we propose a new single network model for interactive segmentation that fully utilizes all available past information to refine the segmentation of follow-up scans. In the first segmentation round, our model takes 3D volumes of medical images from two-time points (target and reference) as concatenated slices with the additional reference time point segmentation as a guide to segment the target scan. In subsequent segmentation refinement rounds, user feedback in the form of scribbles that correct the segmentation and the target's previous segmentation results are additionally fed into the model. This ensures that the segmentation information from previous refinement rounds is retained. Experimental results on our in-house multiclass longitudinal COVID-19 dataset show that the proposed model outperforms its static version and can assist in localizing COVID-19 infections in patient's follow-up scans.
Segmentation-based object categorization
Region growing
Cite
Citations (0)
Model fusion can effectively improve the effect of model prediction, but it will bring about an increase in time. In this paper, the dual-stage progressive knowledge distillation is improved in combination with multi-teacher knowledge distillation technology. A simple and effective multi-teacher's Softtarget integration method is proposed in multi-teacher network knowledge distillation. Improve the guiding role of excellent models in knowledge distillation. Dual-stage progressive knowledge distillation is a method for small sample knowledge distillation. A progressive network grafting method is used to realize knowledge distillation in a small sample environment. In the first step, the student blocks are grafted one by one onto the teacher network and intertwined with other teacher blocks for training, and the training process only updates the parameters of the grafted blocks. In the second step, the trained student blocks are grafted onto the teacher network in turn, so that the learned student blocks adapt to each other and finally replace the teacher network to obtain a lighter network structure. Using Softtarget acquired by this method in Dual-stage progressive knowledge distillation instead of Hardtarget training, excellent results were obtained on BreakHis data sets.
Sample (material)
Cite
Citations (4)
Приведены основные схемы получения молодых коньячных дистиллятов на аппаратах периодического действия двойной сгонки (шарантского типа), традиционно применяемые в отечественном коньячном производстве и у классических французских производителей. При помощи математического моделирования проведен анализ и определен выход коньячного дистиллята с заданными кондициями и удельные энергетические затраты при одинаковых начальных условиях, но при различных вариантах перегонки. Произведен расчет кондиций основных получаемых продуктов и промежуточных фракций за один цикл дистилляции, а также за длительную последовательность циклов для учета влияния возвращаемых головных и хвостовых фракций, на процесс последующих перегонок. Установлено, что схемы получения коньячных дистиллятов, традиционно используемые в странах СНГ, и метод дистилляции MARTELL, являются менее эффективными с точки зрения выхода коньячного дистиллята и удельных энергозатрат, чем метод дистилляции коньячных домов REMY MARTIN и HENNESSY. Установлено, что регулирование объемной доли этилового спирта в спирте-сырце коньячном за счет отбора хвостовой фракции при первой перегонке является эффективным способом управления процессом дистилляции, который оказывает влияние на общий выход кондиционного коньячного дистиллята и удельные энергетические затраты. Максимальный выход коньячного дистиллята и минимальные удельные энергозатраты при перегонке виноматериала с объемной долей этилового спирта 10,5 % и получением коньячного дистиллята с объемной долей этилового спирта 70% достигаются в случае начала отбора хвостовой фракции при объемной доле этилового спирта в парах (спиртовом фонаре) в диапазоне 14-16%. Показано, что применяемые различными производителями схемы получения коньячных спиртов имеют потенциал для оптимизации, позволяющий увеличить выход коньячного спирта до 2,4 % и снизить удельные энергозатраты до 5 %. Оптимизация схемы получения коньячных дистиллятов не требует дополнительных капиталовложений и может служить дополнительным источником прибыли без какой-либо модернизации оборудования. The paper describes basic schemes for producing young brandy distillates on double distillation batch machines (charente type) traditionally used in domestic brandy production and by traditional French producers. Mathematical modelling was used to analyze and determine the output of brandy distillate with predetermined parameters and specific energy costs under the same initial conditions, but with different distillation options. We calculated quality parameters of the main resultant products and intermediate fractions during one distillation cycle, as well as during the long cycle sequence to estimate the effect of returned fraction heads and tails on the subsequent distillation process. The analysis established that brandy distillate production schemes traditionally used in the CIS countries and the MARTELL distillation method are less effective in terms of brandy distillate output and specific energy consumption as compared to the REMY MARTIN and HENNESSY cognac distillation method. It was established that control of the volume fraction of ethyl alcohol in the raw brandy alcohol by tail fraction takeoff during the first distillation is an effective way to control the distillation process, which affects the overall output of conditioned brandy distillate and specific energy costs. The maximum brandy distillate output and the minimum specific energy consumption during base wine distillation with volume fraction of ethyl alcohol at 10.5 % and brandy distillate production with volume fraction of ethyl alcohol at 70 % is achieved when the tail fraction takeoff begins with ethyl alcohol volume fraction in pairs (alcohol lamp) within the range of 14-16 %. It is demonstrated that production schemes used by various producers to obtain brandy spirits can be optimized, which would increase the output of brandy spirits by 2.4 % and reduce the specific energy costs by 5 %. Optimization of the cognac distillates production scheme does not require additional investment, and can serve as an additional profit source without any equipment upgrade.
Fraction (chemistry)
Batch distillation
Volume fraction
Continuous distillation
Cite
Citations (0)
This research is aimed to improve the quality of vetiver oils from smallholders in Indonesia by vacuum distillation. The most important parameters of quality mentioned are total vetiverol content and color. It was shown that vetiverol contents could be increased to achieve the required minimum content of 50%. The better the initial sample, the better the distillate obtained. Distillate fractions obeying standard vetiverol content could be obtained with yield of 60%~80%. Although the initial samples were black in color, the distillates had appearance from yellow to reddish brown, as required by the standard, with Gardner scales of color ranging from 10.8 to 14.7. Distillation, however, slightly disturbed the achievement of other parameters including density, acid number and ester number. Lower distillation fractions tend to shift the values of these parameters to out of standards.
Edible oil
Cite
Citations (2)
Solar still
Distilled water
Potable water
Cite
Citations (82)