MMSA
MMSA copied to clipboard
MMSA is a unified framework for Multimodal Sentiment Analysis.
提供的数据集下载链接中的SIMS数据集似乎是未对齐的(unaligned_39.pkl),请问中文数据集有哪些对其工具?
感谢大佬开源的代码和教程,我有几个问题想礼貌请教一下, 1. 我想使用自己的视频数据进行二分类任务,请问特征提取是必须要切分成video clip的形式吗,是否可以整个视频进行提取,如果可以请问label.csv中的clip id改如何填写; 2. 代码中应该如何增加自己数据集的配置(比如config文件中); 3. 最新的模型TETFN,CENET和TSST是没有集成嘛,运行会报错没有这个模块; 再次感谢各位大佬的工作!
文本中的进行2/5/7分类的计算准确率的指标的时候采用的四舍五入到-3~3的整数,然后进行mirco-weight的指标计算,这种计算方式是通用的吗
I noticed that the audio feature length of the MOSI dataset is 5. May I ask how the audio features are extracted for the MOSI dataset.
raise RuntimeError( RuntimeError: Failed to import transformers.models.bert because of the following error (look up to see its traceback): cannot import name '_sentencepiece' from partially initialized module 'sentencepiece' (most likely due...
代码如下: 特征提取: fet = FeatureExtractionTool(config='/data/huangjingwang/SCI/MMSA/cut/config2.json') infile_path=os.path.abspath('/data/huangjingwang/SCI/MMSA/cut/test_output/output_video_002.mp4') outfile_path=os.path.abspath('/data/huangjingwang/SCI/MMSA/cut/test_output/feature_get.pkl') text_path=os.path.abspath("/data/huangjingwang/SCI/MMSA/cut/input.txt") fet.run_single(in_file=infile_path, out_file=outfile_path,text="What is love") 模型训练: config=get_config_regression('mtfn','sims') config['custom_feature'] = '/data/huangjingwang/SCI/MMSA/dataset/SIMS/unaligned_39.pkl' MMSA_run('mtfn', 'sims', gpu_ids=[0], config=config, res_save_dir='/data/huangjingwang/SCI/MMSA/results/SIMS', model_save_dir='/data/huangjingwang/SCI/MMSA/models/SMIS') 模型测试: MMSA_test(config=config, weights_path='/data/huangjingwang/SCI/MMSA/models/SMIS/mtfn-sims.pth', feature_path='/data/huangjingwang/SCI/MMSA/cut/test_output/feature_get.pkl') config2.json内容: { "audio":...
请问对于分类任务中使用的mosei和chsims数据集是对齐的还是未对齐的?
我有一些自己的视频数据集进行测试,已经使用MMSA-FET进行了特征提取(openface提取视频特征),如何使用MMSA_test()进行测试?(主要是config如何配置?weight如何获取?还是说MMSA_test()需要先训练/微调才可以用?) 谢谢!
I couldn't find the KMeans Clustering Method applied in the CENet source code in MMSA, could you please tell me the reason why you abandoned the method?