Skip to content

Latest commit

 

History

History
91 lines (69 loc) · 3.23 KB

Command.md

File metadata and controls

91 lines (69 loc) · 3.23 KB

TrailBlazer: Commands

Here we illustrate how to make your own config under the current implementation. Please use the following commands to run our TrailBlazer. The config structure is detailed here

bin/CmdTrailBlazer.py

python bin/CmdTrailBlazer.py --config ${YOUR_CONFIG_FILEPATH} -mr ${YOUR_MODEL_ROOT}
e.g., python bin/CmdTrailBlazer.py --config config/PerspBR2TL-Tiger.yaml -mr ${YOUR_MODEL_ROOT}

If you feel the flag -mr(--model-root) is combersome, you can create a shell environment variable named ZEROSCOPE_MODEL_ROOT so that flag can be ignored. It's the folder path that contains cerspense/zeroscope_v2_576w

export ZEROSCOPE_MODEL_ROOT=/your/huggingface/model/root/
# Note: then we expect the zeroscope model is here /your/huggingface/model/root/cerspense/zeroscope_v2_576w/

Then it above command now becomes

python bin/CmdTrailBlazer.py --config ${YOUR_CONFIG_FILEPATH}
e.g., python bin/CmdTrailBlazer.py --config config/Main/PerspBR2TL-Tiger.yaml

For your convenience, you could run the following command to get the results from all config yamls in the given folder:

python bin/CmdTrailBlazer.py --config ${YOUR_CONFIG_FOLDER}
# say, you want to execute all yaml files in the provided config folder
e.g., python bin/CmdTrailBlazer.py --config config/Main/

bin/CmdTrailBlazerMulti.py

As outlined in the paper, the individual objects are initially generated independently before being integrated. For example, if one wishes to guide both a dog and a cat based on the prompt "a dog and a cat running in the park," each needs to be processed separately using our CmdTrailBlazer command:

python bin/CmdTrailBlazer.py --config config/Multi/MultiSubject-Dog.yaml
python bin/CmdTrailBlazer.py --config config/Multi/MultiSubject-Cat.yaml

Upon execution, you will receive the reconstructed video as usual. Additionally, a .pt file will be generated containing the bounding box at each frame and the latent vectors at each step. For example:

/tmp/TrailBlazer/MultiSubject-Dog.0000.pt
/tmp/TrailBlazer/MultiSubject-Cat.0000.pt

After placing these .pt files in the "subjects" key within the configuration, proceed to execute the distinct command CmdTrailBlazerMulti to obtain the final result.

python bin/CmdTrailBlazerMulti.py --config config/Multi/MultiSubjects.yaml

bin/CmdPeekaboo.py

We've managed the entry point of Peekaboo within our executable script CmdPeekaboo, enhancing compatibility with our configuration file in their implementation. Please checkout the page Peekaboo.md for more information about the reproducibility, and the Config.md for the configuration design about Peekaboo.

python bin/CmdPeekaboo.py --config config/config.yaml

We notice that the full precision is used in Peekaboo implementation (e.g., torch.float (See src/generate.py#L76)) that goes beyond our VRAM to run the batch execution (e.g., folder path in --config flag). If you want to run all config under specific folder. Here is an alternative way:

for f in config/*.yaml; do python bin/CmdPeekaboo.py --config $f; done

Get some coffee after executing it :)