File size: 1,657 Bytes
2006222 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 |
# nvidia-smi | grep 'python' | awk '{ print $5 }' | xargs -n1 kill -9
timestamp=$(date +%y%m%d_%H%M%S)
DATASET="nuscenes" #"kitti/vkitti/bdd100k/..."
DATASET_PATH="/network/scratch/a/anthony.gosselin"
NAME="${DATASET}_box_predict_2" #"${DATASET}_box_predict_${timestamp}"
OUT_DIR="/network/scratch/a/anthony.gosselin/Results/ctrlv/${NAME}"
mkdir -p $OUT_DIR
PROJECT_NAME='ctrl_v'
SCRIPT_PATH=$0
SAVE_SCRIPT_PATH="${OUT_DIR}/train_scripts.sh"
cp $SCRIPT_PATH $SAVE_SCRIPT_PATH
echo "Saved script to ${SAVE_SCRIPT_PATH}"
CUDA_LAUNCH_BLOCKING=1 accelerate launch tools/train_video_diffusion.py \
--run_name $NAME \
--data_root $DATASET_PATH \
--project_name $PROJECT_NAME \
--pretrained_model_name_or_path stabilityai/stable-video-diffusion-img2vid-xt \
--output_dir $OUT_DIR \
--variant fp16 \
--dataset_name $DATASET \
--train_batch_size 1 \
--learning_rate 5e-6 \
--checkpoints_total_limit 2 \
--checkpointing_steps 200 \
--gradient_accumulation_steps 5 \
--validation_steps 100 \
--enable_gradient_checkpointing \
--lr_scheduler constant \
--report_to wandb \
--seed 1234 \
--mixed_precision fp16 \
--clip_length 25 \
--min_guidance_scale 3 \
--max_guidance_scale 7 \
--noise_aug_strength 0.01 \
--bbox_dropout_prob 0.1 \
--conditioning_dropout_prob 0.0 \
--num_demo_samples 10 \
--backprop_temporal_blocks_start_iter -1 \
--num_train_epochs 2 \
--predict_bbox \
--num_inference_steps 30 \
--resume_from_checkpoint latest \
--num_cond_bbox_frames 3 \
--wandb_entity chris-pal \
--fps 7
# --if_last_frame_trajectory |