Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Why I can't reproduce your results about | ResNet34-VTN| RGB | UCF101 | 90.27%|

See original GitHub issue

I have two 2080Ti GPUs. I run the commad python main.py --dataset ucf101_1 --model resnet34_vtn -b16 --lr 1e-5 --seq 16 --st 2 But the highest Video@1 is only 85% instead of 90.27% as you reported. So I wonder what might be wrong with my training. Or I overlooked some of the training details you used？ Thanks for your wonderful project and it help me a lot whether I am at Intel or school. Looking forward to your guidance!

Issue Analytics

State:
Created 4 years ago
Comments:5 (2 by maintainers)

Top GitHub Comments

2reactions

vadimadrcommented, Mar 19, 2020

Hi! In order to reproduce the 90%+ result on UCF you need to start with model pretrained on Kinetics.

1reaction

minhhoangbuicommented, Mar 19, 2020

@JinYAnGHe Do you succeed to reproduce the result? I fail it too I suspect they use some tricks to train the model since when I use their model to test, I got the result they reported @vadimadr Can you help us here?

Top Results From Across the Web

Why I can't reproduce your results about | ResNet34-VTN| RGB

I have two 2080Ti GPUs. I run the commad python main.py --dataset ucf101_1 --model resnet34_vtn -b16 --lr 1e-5 --seq 16 --st 2

UCF101 Benchmark (Action Recognition) - Papers With Code

Rank Model 3‑fold Accuracy Year 1 SMART 98.64 2020 2 OmniSource (SlowOnly‑8x8‑R101‑RGB + I3D‑Flow) 98.6 2020 3 PERF‑Net (multi‑distilled S3D) 98.6 2020

Extreme Low-Resolution Activity Recognition with Confident ...

Experimental results demonstrate that, the proposed method can effectively improve the accuracy of eLR activity recognition and.

Can Spatiotemporal 3D CNNs Retrace the History of 2D ...

The purpose of this study is to determine whether current video datasets have sufficient data for training very deep convolutional neural networks (CNNs) ......

Validation loss and accuracy for a Resnet-34 on UCF-101 ...

Bayesian Bits employs a novel decomposition of the quantization operation, which sequentially considers doubling the bit width. At each new bit width, the...