#10 More details about MSVD on zero-example video retrieval
Opened this issue · 12 comments
In the paper, "So we assess the models previously trained on MSR-VTT using the MSVD test set" refers to training with the entire data set of MSRVTT, and testing the model with the test set of MSVD（670 vidoes）? Or the test model is to use the entire MSVD dataset（1970 vidoes） as a test set？
Sorry for the confusion. Indeed, we directly used the Dual encoding model in Table 1 without any finetune, so the model was trained in the training set (6,513 video clips) of MSRVTT. Besides, we evaluated the model on the test set (670 video clips) of MSVD.
Thanks!After I read your other paper 《Word2VisualVec: Image and Video to Sentence Matching by Visual Feature Prediction》，I have some questions, and I wrote on danieljf24/w2vv#11
I really hope you can tell me more details.Thank you!
I would like to ask you a question: in the MSVD data set, especially in the test phase, do you evaluate all sentences, or just randomly select 5 sentences from the MSVD for evaluation?
On MSVD, we tested on the test set of 670 video clips, and using all the corresponding sentences.
It is worth noting that in order to evaluate the generalization ability of our proposed model, the results on MVSD were obtained using the model trained in the training set (6,513 video clips) of MSRVTT.
In the previous version of our w2vv paper, in Table 5, we used all the corresponding sentences instead of randomly sampled 5 sentences for each test video ( Results using data partition from Xu et al. [40]). For the results using data partition from Otani et al. [24], we used 5 sentences for each test video provided by Otani et al, while all the training sentences. 1,200, 100 and 670 video clips for training, validation, and test.
The MSVD data with extracted video features have been updated. https://github.com/danieljf24/dual_encoding#dual-encoding-on-msvd
I am sorry to bother you, can you share the entails of code in MSVD？
Thank you!!
谢谢！！
Can we communicate with each other? Because I'm also from China.
I am sorry to bother you, can you share the entails of code in MSVD？
The corresponding code has also been released.