search

chenjingong / DN-ReID

[CVPR2024]Day-Night Cross-domain Vehicle Re-identification
https://openaccess.thecvf.com/content/CVPR2024/papers/Li_Day-Night_Cross-domain_Vehicle_Re-identification_CVPR_2024_paper.pdf
17 stars 1 forks source link

batch size #4

Open xzb666 opened 2 hours ago

xzb666 commented 2 hours ago

Hello, I used your code with a batch size of 8 during the training process, but the test results are slightly lower than those reported in your paper. May I ask if you trained with a batch size of 8 or 64?

chenjingong commented 1 hour ago

batchsize is 64. 8 is the number of classification in a batchsize.

---- Replied Message ---- | From | @.> | | Date | 10/11/2024 16:15 | | To | @.> | | Cc | @.***> | | Subject | [chenjingong/DN-ReID] batch size (Issue #4) |

Hello, I used your code with a batch size of 8 during the training process, but the test results are slightly lower than those reported in your paper. May I ask if you trained with a batch size of 8 or 64?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

xzb666 commented 1 hour ago

parser.add_argument('--batch-size', default=8, type=int, metavar='B', help='training batch size') Is switching the batch size to 64 sufficient, or do I need to make other modifications as well?

chenjingong commented 1 hour ago

I think you can use the code directly after you change the datepath

---- Replied Message ---- | From | @.> | | Date | 10/11/2024 16:27 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [chenjingong/DN-ReID] batch size (Issue #4) |

Is switching the batch size to 64 sufficient, or do I need to make other modifications as well?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>