Vision-Language Models (VLMs) have achieved impressive performance across various tasks, but often struggle to apply newly introduced visual concepts during inference. A common failure pattern is what we call Mixing Things Up: VLMs frequently confuse concept names, resulting in vague descriptions and failure to ground the concept correctly. Existing approaches mainly address person-related concepts through text prompts or tokenizer modifications. However, VLMs still miss or misinterpret untrained visual concepts, underscoring the need to learn new concepts directly from visual input, without relying on prior textual injection. To overcome these limitations, we propose BISCUIT (Basis-aligned Inference through Structured Concept Unification and Identification-aware Tuning), a two-step training method. Step I proposes a dual-stream structure-aware vision encoder that fuses RGB and edge-based embeddings within a shared basis space to enhance concept recognition. Step II enhances generation quality through identification-aware tuning, which encourages alignment between the generated text and the newly introduced visual concepts. Existing methods mainly focus on person concepts and lack comprehensive evaluation across diverse visual categories. We further propose a benchmark BiscuitVQA to evaluate VLMs performance on recognizing and applying novel image-introduced concepts across diverse concept types and task types, including real people, cartoons, animals, and symbolic content. We apply BISCUIT to LLaVA-1.5 and Qwen2.5-VL, achieving competitive results among open-source models and narrowing the gap to Gemini-2.5 and GPT-4o. Interestingly, our BISCUIT maintains strong generalization, showing minimal degradation on other downstream tasks.

Time synchronization is a key technology in time-sensitive networking (TSN) to support deterministic data transmission with bounded latency, low delay jitter, and zero congestion loss guarantee. TSN uses IEEE 802. IAS protocol to achieve time synchronization. The asymmetry of the propagation link, clock drift, and limited node clock frequency resolution limit the TSN time synchronization performance improvement. In this paper, we propose a scheme for enhancing the time synchronization performance of TSN and design a method to enhance the TSN time synchronization accuracy by precisely measuring the deviation of the start frame delimiter (SFD) identification signal of the frame initiator from the local received clock signal, and taking the measured value as the compensation amount. Furthermore, we introduce a time correction algorithm in the compensation process. The corresponding development implementation is completed in FPGA, and a test environment is constructed for experimental verification. The experimental results obtain a time synchronization accuracy of 7.5ns, and the synchronization accuracy is even better than 5ns after introducing the time correction algorithm.

This work demonstrates the use of depth panoramas in the construction of detailed 3D models of extended environments. The paper describes an approach to the acquisition of such panoramic images using a robotic platform that collects sequences of depth images with a commodity depth sensor. These sequences are stitched into panoramic images that efficiently capture scene information while reducing noise in the captured imagery. Scene structure is extracted from the panoramas and used to register a collection of panoramas to obtain coverage of extended areas. The presented approach maintains fine geometry detail in 3D reconstructions, uses a modest amount of memory, and enjoys excellent scaling characteristics across large environments.