search

lhoyer / MIC

[CVPR23] Official Implementation of MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation
261 stars 40 forks source link

use check point for pre_train on custom dataset #58

Closed TahereSaidi closed 4 months ago

TahereSaidi commented 1 year ago

Hello. I want to use your algorithm But on my dataset. I want to use MIC for object detection. Also, I don't want to make the model from scratch and I want to use the provided checkpoints to fine-tune . But I don't know how to do this. Do I need to do the commands in [INSTALL.md] for fine-tuning? it will be favor to me to resolve my problem

krumo commented 1 year ago

Hi, sorry for the late reply. As the detection part is implemented based on maskrcnn-benchmark, I would suggest you follow their example like e2e_faster_rcnn_R_101_FPN_1x_caffe2.yaml to finetune a pretrained model on a dataset. Specifically, I believe you can specify the MODEL.WEIGHT to a path link to the provided checkpoints and start training on your custom data. For fine-tuning, you still have to follow the steps in [INSTALL.md] to configure the environment to run the code, but you don't have to run the Cityscapes-related commands.