It's a teamwork of @rederxz, @poseidonchan and @ethanyang2000. This project can covert any input images to traditional Chinese poems. We constructed a data set containing pictures and Chinesepoems then built a model based on ResNet to extract keywords from pictures and a character-wise model to generate Chinese poems via RNN.
Detailed information can be found in the report
- pic2poem is the finalized codes. You can add the test samples(images) you want in the
test/
and run this file directly, then you will get the generated poems. - ImageCollector is the code of the web crawler.
- PreProcess is used for the pre-process of the poems.
- ResNet contains the codes for the training of the extractor.
- poem_generator contains the codes for the poem generator.
- all
.npy
and.pt
files are image-poem labels(for our constructed dataset) and trained parameters of the model respectively.