trap 'kill 0' SIGINT | |
export MODEL_PATH="./pretrained/CogVideoX-5b-I2V" | |
processed_data_name=$1 | |
ckpt_steps=500 | |
ckpt_dir=./out/EPiC_pretrained | |
ckpt_file=checkpoint-${ckpt_steps}.pt | |
ckpt_path=${ckpt_dir}/${ckpt_file} | |
video_root_dir="./data/${processed_data_name}" | |
out_dir=${ckpt_dir}/test/${ckpt_steps}_${processed_data_name} | |
CUDA_VISIBLE_DEVICES=0 python inference/cli_demo_camera_i2v_pcd.py \ | |
--video_root_dir $video_root_dir \ | |
--base_model_path $MODEL_PATH \ | |
--controlnet_model_path $ckpt_path \ | |
--output_path "${out_dir}" \ | |
--start_camera_idx 0 \ | |
--end_camera_idx 8 \ | |
--controlnet_weights 1.0 \ | |
--controlnet_guidance_start 0.0 \ | |
--controlnet_guidance_end 0.4 \ | |
--controlnet_input_channels 3 \ | |
--controlnet_transformer_num_attn_heads 4 \ | |
--controlnet_transformer_attention_head_dim 64 \ | |
--controlnet_transformer_out_proj_dim_factor 64 \ | |
--controlnet_transformer_out_proj_dim_zero_init \ | |
--vae_channels 16 \ | |
--num_frames 49 \ | |
--controlnet_transformer_num_layers 8 \ | |
--infer_with_mask \ | |
--pool_style 'max' \ | |
--seed 1 | |