ducminhkhoi / ifs-rcnn Goto Github PK
View Code? Open in Web Editor NEWiFS-RCNN: An Incremental Few-shot Instance Segmenter (CVPR 2022)
License: Apache License 2.0
iFS-RCNN: An Incremental Few-shot Instance Segmenter (CVPR 2022)
License: Apache License 2.0
Hi @ducminhkhoi , can you please let me know where exactly the loss is calculated in Equation 1?. I am checking through file in (iFSCRCNN-> modeling -> roi_heads -> fastrcnn.py), but I didn't get any variance matrix for classifier weights that mentioned in equation.
Hi @ducminhkhoi ,
In the base training stage of train.sh, there is no shot number. It is different in the paper's experiment, as I read there are 1,5,10 shots in the bases classes training (Table 2).
Do you train on 1,5,10 shots for bases classes, or you train without considering "shot" for base classes?
arch=50
network=mask_rcnn
num_gpus=8
dataset=coco
detection_folder=${dataset^^}-detection
# standard softmax Mask-RCNN
suffix=''
python3 -m tools.train_net --num-gpus $num_gpus --config-file configs/${detection_folder}/${network}/${network}_R_${arch}_FPN_base${suffix}.yaml
Thank you.
Hello @ducminhkhoi,
Firstly, thank you for your great work!!
Could you you provide the pre-trained model(weights) for "configs/COCO-detection/mask_rcnn/mask_rcnn_R_50_FPN_base.yaml" as well??
I want to do only fine-tuning on my custom data (as novel classes) so it will be very helpful if you can provide pre-trained weights(probably, model_final.pth) which trained on COCO base classes.
Thank you in advance..
Hi @ducminhkhoi , It is a great contribution to few shot incremental object detection problem. I am trying to replicate the results that given in the paper. I am getting following issue.
Thanks
Hello!When I run the bash finetuning.sh ,
## # For test all base + novel classes
python3 -m tools.ckpt_surgery --coco \
--src1 checkpoints/coco/${network}/${network}_R_${arch}_FPN_base${suffix}/model_final.pth \
--src2 checkpoints/coco/${network}/${network}_R_${arch}_FPN_ft_novel_${shot}shot${suffix}${suffix2}/model_final.pth \
--method combine \
--save-dir checkpoints/coco/${network}/${network}_R_${arch}_FPN_all_final_${shot}shot${suffix}${suffix2}
#
I got the error:
Traceback (most recent call last):
File "/home/ecust/1TSSD/anaconda3/envs/iFS-RCNN/lib/python3.7/runpy.py", line 193, in _run_module_as_main
"main", mod_spec)
File "/home/ecust/1TSSD/anaconda3/envs/iFS-RCNN/lib/python3.7/runpy.py", line 85, in _run_code
exec(code, run_globals)
File "/home/ecust/1TSSD/ycf/iFS-RCNN/tools/ckpt_surgery.py", line 343, in
combine_ckpts(args)
File "/home/ecust/1TSSD/ycf/iFS-RCNN/tools/ckpt_surgery.py", line 140, in combine_ckpts
surgery_loop(args, surgery)
File "/home/ecust/1TSSD/ycf/iFS-RCNN/tools/ckpt_surgery.py", line 236, in surgery_loop
surgery(param_name, True, tar_size_dict[param_name], ckpt, ckpt2)
File "/home/ecust/1TSSD/ycf/iFS-RCNN/tools/ckpt_surgery.py", line 98, in surgery
pretrained_weight = ckpt['model'][weight_name_temp]
KeyError: 'roi_heads.mask_head.predictor.weight'
I want to know why it happened?
Hi anh Khoi,
Em có một vài thắc mắc:
Model này có phải training theo Meta learning không?
Topic này bao gồm few-shot learning. Nếu vậy thì "coco_trainval_novel_1shot" có phải là support set, còn "coco_test_novel" thì là query set phải không anh?
E.g: config file:
TRAIN: ('coco_trainval_novel_1shot',)
TEST: ('coco_test_novel',)
Thanks anh.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.