search

GuangxingHan / FCT

Code for CVPR 2022 Oral paper: 'Few-Shot Object Detection with Fully Cross-Transformer'
72 stars 13 forks source link

AssertionError #3

Open tutu-star opened 1 year ago

tutu-star commented 1 year ago

How can I get 'pvt_v2_b2_li.pth' ? I can't find this file. image

GuangxingHan commented 1 year ago

The pre-trained PVTv2 models are here https://github.com/whai362/PVT/tree/v2/classification#model-zoo

tutu-star commented 1 year ago

Hello, I have a question. After downloading your pre training model, I want to know how to use your code for testing.

GuangxingHan commented 1 year ago

The pre-trained model is after the second step training (two-branch based model over base classes). You can directly use the script of few-shot fine-tuning for evaluation. For example, on COCO dataset, you can use the following script

sh scripts/two_branch_few_shot_finetuning_coco_pvt_v2_b2_li.sh
tutu-star commented 1 year ago

Hello, when I use my own dataset for evaluation, the following problems are reported. How can I solve them? image

GuangxingHan commented 1 year ago

It seems that the direct cause is that dataset_name is not an attribute of meta. You can follow the coco_evaluation or pascal_voc_evaluation to implement the evaluation script for your dataset. Actually according to Line 42 in pascal_voc_evaluation, we use meta.dirname rather than meta.dataset_name.