Dear Organizers,
In this competition, the model size is restricted to a maximum of 20MB, and the computational cost is limited to 120 GFLOPs per video during the inference/test stage. However, which standards are used to calculate these metrics? Is there any official code available for this calculation?
Best regards
Posted by: sunguwei @ Feb. 7, 2025, 2:51 p.m.Thanks for your question, we will provide the measurements for parameters and gflops
Posted by: Xin.Li @ Feb. 11, 2025, 12:22 p.m.Furthermore, I have a question regarding the model size. Typically, we assess efficiency based on the number of parameters in the model. However, given the model size limit of less than 20MB, it appears that we may need to restrict the model to fewer than 10 million parameters. This seems to create a notable discrepancy when compared to the GFlop limit. Could you please clarify these constraints?
Posted by: cydiachen @ Feb. 13, 2025, 7:11 a.m.Thanks for your question. We have remove the limitation for model size. Thanks.
Posted by: Xin.Li @ Feb. 17, 2025, 12:53 p.m.