Closed yzfly closed 3 years ago
liming-ai
commented
3 years ago Can you share your training settings for ResNet50-TSM? I use following training settings and only get testing Prec@1 47.856.
--arch "MS" --num_segments 8 --mode 1 --gd 200 --lr 0.02 --lr_steps 20 30 --epochs 40 -b 32 -i 1 -j 4 --dropout 0.5
Testing Results: Prec@1 47.856 Prec@5 78.021 Loss 2.20553 Time 0.1840
hi, I have the same question, have you resolve this? Could you please tell me how to fix it?
yzfly
commented
3 years ago Can you share your training settings for ResNet50-TSM? I use following training settings and only get testing Prec@1 47.856. --arch "MS" --num_segments 8 --mode 1 --gd 200 --lr 0.02 --lr_steps 20 30 --epochs 40 -b 32 -i 1 -j 4 --dropout 0.5 Testing Results: Prec@1 47.856 Prec@5 78.021 Loss 2.20553 Time 0.1840
hi, I have the same question, have you resolve this? Could you please tell me how to fix it?
No, I also wander how to reproduce the results.
TimandXiyu
commented
3 years ago Can you share your training settings for ResNet50-TSM? I use following training settings and only get testing Prec@1 47.856. --arch "MS" --num_segments 8 --mode 1 --gd 200 --lr 0.02 --lr_steps 20 30 --epochs 40 -b 32 -i 1 -j 4 --dropout 0.5 Testing Results: Prec@1 47.856 Prec@5 78.021 Loss 2.20553 Time 0.1840
hi, I have the same question, have you resolve this? Could you please tell me how to fix it?
No, I also wander how to reproduce the results.
I don't think this version of the code could ever match up with the paper's statement. It barely offers any accuracy improvement on minkinetics and hmdb51. I am very doubtful this is the final version of the working code.
arunos728
commented
3 years ago According to the results, it seems that 'Spatial Correlation Sampler' does not work. We just have provided another matching layer (Matching_layer_mm in resnet_TSM.py) which consists of matrix multiplication & index rearrangement. Originally, the output of (Matching_layer_scs) and (Matching_layer_mm) should be the same. You can check out the result of SpatialCorrelationSampler_and_MatMul.ipynb. If the tensor outputs of two matching layers are not the same, you should re-install Spatial Correlation Sampler based on environment or just use Matching_layer_mm (line 310 instead of line 309 in resnet_TSM.py). We also provide the instructions for installing Spatial Correlation Sampler'.
Can you share your training settings for ResNet50-TSM? I use following training settings and only get testing Prec@1 47.856.
--arch "MS" --num_segments 8 --mode 1 --gd 200 --lr 0.02 --lr_steps 20 30 --epochs 40 -b 32 -i 1 -j 4 --dropout 0.5
Testing Results: Prec@1 47.856 Prec@5 78.021 Loss 2.20553 Time 0.1840