updates

training/mochi-1/prepare_dataset.sh

@@ -1,49 +1,9 @@
 #!/bin/bash
 
-MODEL_ID="genmo/mochi-1-preview"
+GPU_ID=0
+VIDEO_DIR=/home/sayak/cogvideox-factory/video-dataset-disney-organized
+OUTPUT_DIR=videos_prepared
 
-NUM_GPUS=1
+python trim_and_crop_videos.py $VIDEO_DIR $OUTPUT_DIR --num_frames=37 --resolution=480x848 --force_upsample
 
-# For more details on the expected data format, please refer to the README.
-DATA_ROOT="/home/sayak/cogvideox-factory/video-dataset-disney"  # This needs to be the path to the base directory where your videos are located.
-CAPTION_COLUMN="prompt.txt"
-VIDEO_COLUMN="videos.txt"
-OUTPUT_DIR="/home/sayak/cogvideox-factory/video-dataset-disney/mochi-1/preprocessed-dataset"
-HEIGHT_BUCKETS="480"
-WIDTH_BUCKETS="848"
-FRAME_BUCKETS="85"
-MAX_NUM_FRAMES="85"
-MAX_SEQUENCE_LENGTH=256
-TARGET_FPS=30
-BATCH_SIZE=4
-DTYPE=fp32
-
-# To create a folder-style dataset structure without pre-encoding videos and captions
-# For Image-to-Video finetuning, make sure to pass `--save_image_latents`
-CMD_WITHOUT_PRE_ENCODING="\
-  torchrun --nproc_per_node=$NUM_GPUS \
-    prepare_dataset.py \
-      --model_id $MODEL_ID \
-      --data_root $DATA_ROOT \
-      --caption_column $CAPTION_COLUMN \
-      --video_column $VIDEO_COLUMN \
-      --output_dir $OUTPUT_DIR \
-      --height_buckets $HEIGHT_BUCKETS \
-      --width_buckets $WIDTH_BUCKETS \
-      --frame_buckets $FRAME_BUCKETS \
-      --max_num_frames $MAX_NUM_FRAMES \
-      --max_sequence_length $MAX_SEQUENCE_LENGTH \
-      --target_fps $TARGET_FPS \
-      --batch_size $BATCH_SIZE \
-      --use_slicing \
-      --dtype $DTYPE
-"
-
-CMD_WITH_PRE_ENCODING="$CMD_WITHOUT_PRE_ENCODING --save_latents_and_embeddings"
-
-# Select which you'd like to run
-CMD=$CMD_WITH_PRE_ENCODING
-
-echo "===== Running \`$CMD\` ====="
-eval $CMD
-echo -ne "===== Finished running script =====\n"
+CUDA_VISIBLE_DEVICES=$GPU_ID python embed.py $OUTPUT_DIR --shape=37x480x848