In this, now de-facto standard, setting, an algorithm is provided with several object masks in the first frame and has to output pixel-perfect masks for these objects for the remainder of the video, together …

In this paper, utilizing Taylor expansion, we demonstrate that numerical quadrature causes inevitable approximation error in NeRF integrals due to ignoring the parameter associated with the Lagrange …

Afterwards, we train an instance segmentation model with the univocal points and equivocal points labeled by GLP, and then employ it to assign pseudo labels for the remainder of the unlabeled points.

The remainder of the paper is organized as follows. Sec-tion 2 reviews the related work. Section 3 introduces the proposed methods. Section 4 describes the experimental setup and the achieved …

In this paper we propose a new parameter-efficient fine-tuning method Gradient-based Parameter Selection (GPS) demonstrating that only tuning a few selected parameters from the pre-trained …

The remainder of this section is organized as follows: Sec. 3.1, 3.2, and 3.4 focus on improving local unsuper-vised learning, while Sec. 3.3 and 3.5 targets challenges in federated learning.

To be specific, their models compute a set of heterogeneous embedding vectors for a given sample using self-attention layers on top of vi-sual or textual features; such a set of embedding vectors is …

In the remainder of this section, we de-scribe the re-training of ControlNet for TIR image synthesis and the training of the reference approach for semantic seg-mentation in more detail.

The remainder of the paper is organized as follows. Sec-tion 2 reviews some related research work. Section 3 illus-trates the proposed method in detail. Section 4 elaborates on the experimental results …

In the original Siamese tracker [1], the object template is initialized in the first frame and then kept fixed during the remainder of the video. However, appearance changes are often large and failing to …