IIT-PAVIS / Look_Around_And_Learn

Code for the paper titled "Look Around 👀 and Learn 🎓: Self-Training Object Detection by Exploration" accepted at ECCV2024
MIT License
9 stars 0 forks source link

reproduce of original MaskRCNN model #5

Open yuanxuanS opened 2 weeks ago

yuanxuanS commented 2 weeks ago

When I infer with off-the-shelf MaskRCNN model, it shows bad prediction like this: 微信图片_20241029161922 I set the test score threshold to 0.9. Did you get this result as well?

yuanxuanS commented 2 weeks ago

I used detection_args of maskrcnn. And its config is: `# @package group config_file: ${base_dir}/third_parties/detectron2/configs/quick_schedules/mask_rcnn_R_50_FPN_inference_acc_test.yaml confidence_threshold: 0.9 MODEL: ROI_HEADS: NMS_THRESH_TEST: 0.9

`

morpheus1820 commented 2 weeks ago

Hi, this is the detectron2 configuration file we used,specified inside train_policy.yaml: https://github.com/IIT-PAVIS/Look_Around_And_Learn/blob/e939597aac5de4c8a7f173bbf728cd2dc98d5ec8/confs/detectron_args/maskrcnn.yaml