Could you release all the project code including training and inference stage.

[yangt1013]

Thanks a lot for your code. could you release all the project contents, including training models, training files, Table 2 and Table 3 model's files. there are only generate dataset instruction.

[lorebianchi98]

Thank you for your interest in our work!

To clarify, while we've shared the training and validation datasets, we did not train any models as part of this project. All the methods evaluated in our work are based on pre-trained models. For further details, including checkpoints and installation instructions, please refer to the official repositories of those models.

You can find the scripts used to generate predictions on the FG-OVD benchmark suite here, which will allow you to evaluate the pre-trained models against our benchmarks. Afterward, use this script to calculate the mAP values presented in Table 2 and Table 3.

I hope this clears up any confusion, and feel free to reach out if you have any more questions!

[yangt1013]

Thanks for your reply. About some another problems. Could you provide the details how to generate dataset step by step. I met some problems with following link https://huggingface.co/OpenAssistant/oasst-sft-6-llama-30b-xor/blob/main/README.md Did you download the LLM weights directly to the captions_generation/models or install by the link instructions.

[lorebianchi98]

The first step is to obtain the Llama 2 model weights. You'll first need to request them here. Once approved, follow the instructions provided in the README of the link you shared. After setting everything up, place the downloaded model weights into the _captions_generation/models_ directory.