train_and_eval.md

Training Details

We train FoundationSSC for 25 epochs on 4 NVIDIA 4090 GPUs, with a batch size of 4. It approximately consumes 22GB of GPU memory on each GPU during the training phase. Before starting training, please download the required pretrained checkpoints, including FoundationStereo and EdgeNeXt-Small, and place them in the ckpts directory as follows:

ckpts/
├── 23-51-11
│   ├── cfg.yaml
│   └── model_best_bp2.pth
└── edgenext_small
    └── model.safetensors

Train

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py \
--config_path configs/FoundationSSC-SemanticKITTI.py \
--log_folder FoundationSSC-SemanticKITTI \
--log_every_n_steps 50

The training logs and checkpoints will be saved under the log_folder.

Evaluation

Downloading the checkpoints from the model zoo and putting them under the ckpts folder.

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py \
--eval --ckpt_path ./ckpts/FoundationSSC-SemanticKITTI.ckpt \
--config_path configs/FoundationSSC-SemanticKITTI.py \
--log_folder FoundationSSC-SemanticKITTI-eval \
--log_every_n_steps 50

Evaluation with Saving the Results

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py \
--eval --ckpt_path ./ckpts/FoundationSSC-SemanticKITTI.ckpt \
--config_path configs/FoundationSSC-SemanticKITTI.py \
--log_folder FoundationSSC-SemanticKITTI-eval \
--log_every_n_steps 50 --save_path pred

The results will be saved into the save_path.

Submission

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py \
--eval --ckpt_path ./ckpts/FoundationSSC-SemanticKITTI.ckpt \
--config_path configs/FoundationSSC-SemanticKITTI.py \
--log_folder FoundationSSC-SemanticKITTI-eval \
--log_every_n_steps 50 --save_path pred --test_mapping