Fast-BEV
Fast-BEV copied to clipboard
Sense-GVT
Question concerning temporal fusion
I understand that Fast-BEV uses 4 consecutive frames to improve performance. Do you correct the BEV features of the past frames before temporal fusion (concatenation)? I would naively think that some correction might be necessary to compensate for the ego-car motion between frames. Could you please guide me where this correction happens in the code? Is it applied for every model or just for the larger ones?
I don't think so since the data preparation doesn't involve can_bus info. Correct me if I am wrong, I am looking for the same point as well.
我不要这么认为，因为数据准备不涉及can_bus信息。如果我错了，请纠正我，我也在寻找同样的观点。
Do you understand now？I don't think it's aligned, it's just concat
I understand now.
Sent from Gmail Mobile
On Mon, Mar 11, 2024 at 2:32 PM loserhou @.***> wrote:
我不要这么认为，因为数据准备不涉及can_bus信息。如果我错了，请纠正我，我也在寻找同样的观点。
Do you understand now？I don't think it's aligned, it's just concat
— Reply to this email directly, view it on GitHub https://github.com/Sense-GVT/Fast-BEV/issues/72#issuecomment-1987717739, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZQ6MVANMEDEKM7EMUDOUUTYXVFZHAVCNFSM6AAAAAA64XKEMGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSOBXG4YTONZTHE . You are receiving this because you commented.Message ID: @.***>