Research map / Label-efficient and robust learning / Recognition and long-tail
Probing Sentiment-Oriented Pre-Training Inspired by Human Sentiment Perception Mechanism
Tinglei Feng, Jiaxuan Liu, Jufeng Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet50 · FI | Acc: 0.670 -> 0.707 | Improved |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet101 · UnBiasedEmo | Acc: 0.823 -> 0.859 | Improved |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet18 · Emotic | Hamming: 0.163 -> 0.145 | Improved |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet50 · Emotic | MacroF1: 0.041 -> 0.043 | Improved |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | Vgg19 · Emotion6 | Hamming: 0.253 -> 0.171 | Improved |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet101 · Emotion6 | MacroF1: 0.825 -> 0.827 | No clear effect |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet50 · Emotion6 | KL: 0.403 -> 0.402 | Hurt |
| Use sentiment-oriented pre-training and amalgamation instead of ImageNet initialization | ResNet50 · Abstract | KL: 0.568 -> 0.517 | Improved |
| Use sentiment-oriented pre-training rather than SimCLR | FI models · FI | Accuracy: 0.627 -> 0.707 | Improved |
| Add ST, HO, and HP feature regularization (ablation) | ResNet50 amalgamation · FI | Acc: 66.95 -> 68.92 | Improved |
| Add logits regularization from the ST model (ablation) | ResNet50 amalgamation · FI | Acc: 66.95 -> 66.38 | Hurt |
| Add logits regularization from the HP model (ablation) | ResNet50 amalgamation · FI | Acc: 66.95 -> 70.20 | Improved |
Closest papers by meaning
- Weakly Supervised Coupled Networks for Visual Sentiment Analysis (CVPR 2018 · similarity 0.89)
- Learning Visual Sentiment Distributions via Augmented Conditional Probability Neural Network (AAAI 2017 · similarity 0.88)
- Visual Sentiment Analysis by Attending on Local Image Regions (AAAI 2017 · similarity 0.87)
- Emotional Attention: A Study of Image Sentiment and Visual Attention (CVPR 2018 · similarity 0.86)
- Multi-Source Domain Adaptation for Visual Sentiment Classification (AAAI 2020 · similarity 0.86)
- Emotion-Aware Human Attention Prediction (CVPR 2019 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
