This paper addresses the limitations of the Contrastive Language-Image Pre-training (CLIP) model’s image encoder and proposes a segmentation model WSSS-ECFE with enhanced CLIP feature extraction, aiming to improve the performance of the Weakly Supervised Semantic Segmentation (WSSS) task. WSSS-ECFE employs the Enhanced Bottleneck module proposed in this paper and adds dynamic residual connection to improve the model’s processing effect on complex scenes. In terms of implementation, the Enhanced Bottleneck module employs the Swish activation function and the Depthwise Separable Convolution to enhance the feature extraction and segmentation capability of the model, and uses multiple attention mechanisms to further optimize the feature representation and segmentation accuracy. The WSSS task on the public datasets PASCAL VOC 2012 and MS COCO 2014 achieves 82.6% and 56.3% mean intersection over union (mIoU), achieving state-of-the-art performance in models with low resource requirements.

Multimodal Aspect-Based Sentiment Analysis aims to detect aspect words in a given coupled sentence image pair and predict their corresponding emotional polarity. In order to improve the performance of MABSA tasks, existing methods have done a lot of work in feature alignment and feature fusion, but they still struggle to effectively address the performance loss caused by process noise and semantic differences between coarse and fine granularity. To alleviate these problems, this paper proposes a two-stage Multi-grained Interaction Guided Fusion Model (MIGFM) for MABSA tasks. In the first stage, this paper designs a new Multi-grained Interactive Split Fusion module (MISF), which first calculates the coarse grained high correlation features among modes, and then promotes the generation of fine grained features through the adaptive interaction mechanism to achieve efficient information fusion between modes. In the second stage, this paper designs a Multi-level Interactive Guidance Attention module (MIGA), which further refines the semantic enhancement features by designing a Sliding Window mechanism, and greatly suppresses the impact of low correlation information on the final results. Extensive experiments conducted on two benchmark datasets have shown that the proposed model outperforms state-of-the-art models in MABSA tasks.

The proposed method in this paper proposes an end-to-end unsupervised semantic segmentation architecture DMSA based on four loss functions. The framework uses Atrous Spatial Pyramid Pooling (ASPP) module to enhance feature extraction. At the same time, a dynamic dilation strategy is designed to better capture multi-scale context information. Secondly, a Pixel-Adaptive Refinement (PAR) module is introduced, which can adaptively refine the initial pseudo labels after feature fusion to obtain high quality pseudo labels. Experiments show that the proposed DSMA framework is superior to the existing methods on the saliency dataset. On the COCO 80 dataset, the MIoU is improved by 2.0, and the accuracy is improved by 5.39. On the Pascal VOC 2012 Augmented dataset, the MIoU is improved by 4.9, and the accuracy is improved by 3.4. In addition, the convergence speed of the model is also greatly improved after the introduction of the PAR module.