Request for Access to Pre-trained Model

[NamHV05]

Hello Mr. Sun,

We are a group of Computer Science students in a developing country, currently working on a project that relies heavily on the impressive work shared in your repository.

We have successfully trained the model on 2 out of the 5 datasets specified, but due to limited budget and computational resources, we are unable to continue training on the remaining datasets. To move forward with the next steps in our project, we kindly request access to the pre-trained model.

We assure you that the model will be used strictly for academic purposes, and we will credit your work appropriately in any presentations or reports we produce.

Thank you for considering our request. We look forward to your response.

Best regards,
Nam
nam090505@gmail.com

[feifeiobama]

Sorry, I no longer have the pre-trained model on my server, perhaps you can request the model from some impressive follow-up work (including those published in CVPR).

[NamHV05]

Thank you for your quick response and for pointing us toward the follow-up work—much appreciated.

I have a couple of follow-up questions about the training process:

1. Is it possible to adjust the code so the model loads segments of the datasets, such as two at a time in sequence (e.g., market1501 and cuhksysu, then cuhksysu and dukemtmc, and so on), instead of loading all five at once? Our current setup, with access to a rented GPU server, doesn't have enough RAM to handle all the datasets simultaneously.
2. How long did it take you to train the model on the entire dataset? This would help us better plan our time and resources.

Thanks again for your time and for sharing your work. Looking forward to hearing your thoughts.

[feifeiobama]

1. The current dataset only loads image paths, which should not cost too much RAM (link). The memory bottleneck must be somewhere else.
2. The training is quite fast, I don't remember the exact time, but each stage should take only a few hours. If you validate the model on the initial stages, you can iterate the algorithm twice a day.