jingkang50 / iccv21_scood

The Official Implementation of the ICCV-2021 Paper: Semantically Coherent Out-of-Distribution Detection.

Hi, there! 👋

🎒 I am pursuing my PhD on the topic of visual perception and reasoning in the open world.

🔭 I’m recently focusing on scene graph generation 🕸, vision language models 🧠, and embodied AI 🤖️.

iccv21_scood's Issues

the "sc_label" in train_tin.txt for benchmark cifar100

Thank you for your great job!

We find that the "sc_label" in train_tin.txt for benchmark cifar100 are all "-1", according to my understanding, "sc_label" represents the gt label of the sample corresponding to CIFAR100 data set. Do you think this needs to be modified?

Reproduce UDG

Hi,
I am trying to replicate the results of the paper with SC-OOD CIFAR-100 benchmarks.
Are the results of papers are not the results of averaged performance over a few runs and are the performance of the best model that shows the best accuracy in the test set?

Thanks.

Why in-distribution classification results are so bad?

I tested your model. The model could detect ood images quite well, but when it comes to classifying in-distribution classes. The model produced very bad results with wrong class label. So why in-distribution classification results are so bad?

The volatility of reproducibility results

Thank you for your great job! I've run your code many times(with idf method of ‘udg’ ),but the results have been fluctuating. Your results in Github are expressed by means and standard ,do you think the volatility comes from the randomness of Clustering used in your paper？

Faiss assertion 'err == CUBLAS_STATUS_SUCCESS' failed in void

Faiss assertion 'err == CUBLAS_STATUS_SUCCESS' failed in void faiss::gpu::runMatrixMult(faiss::gpu::Tensor<float, 2, true>&, bool, faiss::gpu::Tensor<T, 2, true>&, bool, faiss::gpu::Tensor<IndexType, 2, true>&, bool, float, float, cublasHandle_t, cudaStream_t) [with AT = float; BT = float; cublasHandle_t = cublasContext*; cudaStream_t = CUstream_st*] at /__w/faiss-wheels/faiss-wheels/faiss/faiss/gpu/utils/MatrixMult-inl.cuh:265; details: cublas failed (13): (512, 256) x (1000, 256)' = (512, 1000) gemm params m 1000 n 512 k 256 trA T trB N lda 256 ldb 256 ldc 1000
[1]    135532 abort (core dumped)  python train.py --config configs/train/cifar10_udg.yml --data_dir data  

I got this error which cost me a ton of time to fix but still failed. I am using cuda 11.4, torch=1.8, 4 A
100 GPUs