wangchen1801 / fpd Goto Github PK
View Code? Open in Web Editor NEWOfficial code of the paper "Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024)"
Home Page: https://arxiv.org/pdf/2401.07629.pdf
Official code of the paper "Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024)"
Home Page: https://arxiv.org/pdf/2401.07629.pdf
To my understanding, the "Average results over multiple runs" are derived from utilizing 30 distinct sample seeds, as demonstrated in the data split. However, upon examining the re-organized data split, I noticed only one seed annotation provided (as indicated here). Consequently, I can only obtain results from a single run. How can I replicate the process to achieve the "Average results over multiple runs"?
Hello, your code has the file transformers.py, may I ask what it does specifically?I can run the code by deleting this file.
Hello, I am reproducing the VOC dataset split 1- fpd_r101_c4_2xb4_voc split 1base training.py with 2 gpus according to your experimental procedure, but after 20000 iterations, the experimental accuracy is only AP50: 0.7950. May I ask if there are any other files that need to be set up and configured?
Hello, may I ask you a question? I am currently in the fpd_r101_c4_2xb4_voc-split_10shot fine tuning phase, with numynover_shots=10 and num_base_shots=10. I am printing ("supports img. shape:", supports img) in the forward train method of the query_spport. py file, and the output is [20,4,224,224]. I am a bit confused, shouldn't it be [20 * 10,4,224,224]? Could you please guide me? Thank you
Hi Zichen Wang,
I hope you are well.
You have done a wonderful job.The idea is great and thanks alot for sharing the code.
I have two custom datasets and I trained the model for one, giving me good mAP but when train on another, it is giving nan losses after 50 iterations.
Both the datasets are relatively similar except for the object sizes, in the first one, object size are small as compared to the standard voc dataset objects. I got results for this dataset.
While in another, object sizes are of medium size. But I am getting nan loss. The bounding boxes are correct.
If you can help or guide me in this, it would be great.
Thanks & Regards!
Sumayya
Sorry to bother you! If I not only focus on querying images, but also want to know how to visualize the heat map of the support set, do I have any relevant support set visualization code? I hope you can reply in your busy schedule, I would be very grateful!
Hi, I have not been able to figure out the exact training process of meta-learning, can you explain it for me. Here is my understanding and some questions:
Take poscal voc dataset as an example:
First of all meta learning is divided into support set and query set, then the pascal base class is divided into multiple tasks, each task has a corresponding support set and query set, what I understand is that first the pictures from the support set are put one by one into the branch below in the graph, then the class prototype is trained, then each task comes up with the class prototype, how is it ultimately consolidated into a class prototype? Then the whole model is trained after going the branch above in the graph with the images from the query set, the query set loss for each task adds up to the meta loss, and then the whole model is updated.
I see that the FFA module query image in your text incorporates a support set of finesse prototypes, so I think I misunderstood the training process of meta-learning. So can you elaborate the whole training process (from which step the meta loss is calculated and what is updated at the same time) using psacl voc as an example.
Hope to get your help, thanks a lot!
Thanks
Hello, when I was executing the inference.py file, I was prompted that "FPD is not in the models registry". I have executed "python setup.py install", but I am still experiencing this error when inference.
Good job! Hello friend, I would like to ask if I can use a single 3090GPU to learn with a learning rate of 0.0025. Sometimes, suddenly, there is an issue. I noticed that the code you provided is 2 * 3090 with an lr of 0.005, while the one provided in the paper is 0.004. May I ask which specific lr you implemented.
I want to continue working on your work, but I encountered some problems during the code reproduction process! In fpd_r101_c4_2xb4_voc-split_10shot fine-tuning, experiments were conducted according to the code you provided. The log file is shown in 10shot. log, and the optimal accuracy NOVEL-CLASSESPLIT1 mAP: 0.672 was achieved in 2400 iterations of testing results. However, upon observing the log file you provided, only 2000 iterations were performed, achieving an accuracy of 0.684. What is the main reason for the difference in experimental results?
10shot.log
Hello, excuse me! I'm having this problem at runtime, but I'm almost following the steps in the readme to run the configuration, so I'd like to ask how to solve this problem?
Traceback (most recent call last): File "train.py", line 18, in <module> import mmfewshot # noqa: F401, F403 File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/__init__.py", line 6, in <module> from .classification import * # noqa: F401, F403 File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/__init__.py", line 2, in <module> from .apis import * # noqa: F401,F403 File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/apis/__init__.py", line 2, in <module> from .inference import (inference_classifier, init_classifier, File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/apis/inference.py", line 14, in <module> from mmfewshot.classification.models import BaseMetricClassifier File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/models/__init__.py", line 5, in <module> from .classifiers import * # noqa: F401,F403 File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/models/classifiers/__init__.py", line 8, in <module> from .maml import MAML File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/models/classifiers/maml.py", line 10, in <module> from mmfewshot.classification.datasets import label_wrapper File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/datasets/__init__.py", line 5, in <module> from .builder import (build_dataloader, build_dataset, File "/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmfewshot/classification/datasets/builder.py", line 7, in <module> from mmcls.datasets.builder import DATASETS, DistributedSampler, worker_init_fn ImportError: cannot import name 'DistributedSampler' from 'mmcls.datasets.builder' (/home/ubuntu/anaconda3/envs/fpd01/lib/python3.8/site-packages/mmcls/datasets/builder.py)
Excuse me, while reading the code in your ffa.py file, there is a line of code “attn.div_(0.5)” in the forward function of the PrototypesAssignment class, which you did not mention in your paper. Could you please help me answer this? Thank you
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.