1. SlideMaster 🎓

The training code is adapted from llama-factory.

To start model training, execute the following script:

cd SlideMaster
bash train.sh

Since LoRA is used for training, a merge stage is required at the end of training:

bash merge.sh

Then, use vllm to start an OpenAI-Compatible Server:

cd ..
cd SlideCoder
vllm serve ../SlideMaster/output/ppt2code_512_w_intro

2. Slide2Code 🖼️

We provide an example in Slide2Code/input, where the folder structure is as follows:

• Slide2Code/input/background: Contains the Backgrounds of the slides.
• Slide2Code/input/images: Contains the Pictures in the ppt.
• Slide2Code/input/design: Contains the Design images.
• Slide2Code/input/origin: Contains the Reference Image.

Slide2Code/Slide2Code.csv serves as our benchmark. Considering copyright issues, we provide the download URL of each slide's pptx file and the corresponding page number.

3. SlideCoder 🛠️

To successfully run the example, you need to execute the following script to run the SlideCoder framework:

cd SlideCoder
bash scripts/main.sh

After running the above code, you will obtain the final code, which needs to be processed through a program. Copy the file run_all_code.py to the generated folder directory, and then run it to get the final ppt file:

cp run_all_code.py $output_path/run_all_code.py
cd $output_path
python run_all_code.py

The generated pptx is located in the output folder.

The execution results of the example can be found in SlideCoder/example_output.

Additionally, block_based.py is the CGSeg algorithm, SlideCoder/db/TBK.txt is the Shape Type knowledge base, and SlideCoder/db/FBK.txt is the Operation Function knowledge base.