Choose the appropriate version based on your system, then install:
bash
1# Example command, replace with the correct version if needed2pip install -e causal_conv1d>=1.1.0
mamba:
Ensure system compatibility and note that this library is modified in the Vim project.
Installation steps:
bash
1# Example command, ensure system compatibility2pip install -e mamba-1p1p1
File Replacement:
Download or update the modified version in the Vim project:
Specifically, replace the mamba-1p1p1 as needed:
bash
1# Assuming the target file is in a specific path within the Vim folder2cp path/to/modified/mamba-1p1p1 path/to/Vim/mamba-1p1p1/
Please adjust the steps based on your system and project needs, ensuring all paths and version numbers are correct.
Dataset
Please download CVUSA, CVACT and VIGOR. You may need to modify the dataset path in "dataloader".
Model Zoo
Dataset
R@1
R@5
R@10
R@1%
Hit
CVUSA
96.19%
98.62%
99.00%
99.52%
-
CVACT_val
87.62%
94.88%
96.06%
98.06%
-
CVACT_test
81.69%
92.42%
94.32%
97.19%
-
VIGOR Same-Area
55.24%
80.75%
76.12%
97.30%
57.43%
VIGOR Cross-Area
19.31%
37.50%
46.03%
86.96%
20.72%
Note
All related results are available at Hugging Face.
Usage
Training
To train our models on the respective datasets, simply run the following scripts:
For CVUSA:
bash train_CVUSA.sh
For CVACT (validation set):
bash train_CVACT.sh
For CVACT (test set):
bash train_CVACT_test.sh
For VIGOR Same-Area:
bash train_VIGOR.sh
For VIGOR Cross-Area:
bash train_VIGOR_cross.sh
These scripts contain all necessary parameters and configurations to train our method on each dataset for 50 epochs.
Evaluation
You should organize the downloaded pre-trained models in the following way:
./result_cvusa/
model_best.pth.tar
checkpoint.pth.tar
./result_cvact/
model_best.pth.tar
checkpoint.pth.tar
./result_vigor/
model_best.pth.tar
checkpoint.pth.tar
./result_vigor_cross/
model_best.pth.tar
checkpoint.pth.tar
Note:
To evaluate the models, simply add the -e option to the corresponding training script:
Modify the following files by adding -e to the command line in each script:
