| Runtag | Org | Describe the LLMs/LVLMs used to generate the captions | Please provide a short description of this run | Additional Details/Comments | Please give this run a priority for inclusion in manual assessments |
|---|---|---|---|---|---|
| LLaVA-NeXT-Video-32B-Qwen + GPT4o | PolySmart | LLaVA-NeXT-Video-32B-Qwen + GPT4o | LLaVA-NeXT-Video-32B-Qwen + GPT4o | LLaVA-NeXT-Video-32B-Qwen + GPT4o | 1 (highest priority) |
| chatGPT_zeroshot_prompt (paper) | DoshishaUzlDfki | GPT-4o | Firstly we generate the step segment timestamps using the GPT-4o API with our zeroshot prompting and then create the step captions using the same LLM with our meta prompting. | There is no big difference between our first 4 runs. So we assign our first 4 runs an arbitrary order except for the 5th run. | 1 (highest priority) |
| mistral_meta_prompt (paper) | DoshishaUzlDfki | Mistral-large-latest | Firstly we generate the step segment timestamps using the Mistral-large-latest API with our meta prompting and then create the step captions using the same LLM with our meta prompting. | There is no big difference between our first 4 runs. So we assign our first 4 runs an arbitrary order except for the 5th run. | 2 |
| mistral_fewshot_prompt (paper) | DoshishaUzlDfki | Mistral-large-latest | Firstly we generate the step segment timestamps using the Mistral-large-latest API with our fewshot prompting and then create the step captions using the same LLM with our meta prompting. | There is no big difference between our first 4 runs. So we assign our first 4 runs an arbitrary order except for the 5th run. | 3 |
| GPT_meta_prompt (paper) | DoshishaUzlDfki | GPT-4o | Firstly we generate the step segment timestamps using the GPT-4o API with our meta prompting and then create the step captions using the same LLM with our meta prompting. | There is no big difference between our first 4 runs. So we assign our first 4 runs an arbitrary order except for the 5th run. | 4 |
| CoSeg_meta_prompt (paper) | DoshishaUzlDfki | Mistral_large_latest | Firstly we generate the step segment timestamps using our trained CoSeg model and then create the step captions using the Mistral-large-latest API with our meta prompting. | We assign it the 5th run since we think its performance is worse than our first 4 runs. | 5 (lowest priority) |