It takes too much time to create data

[arturia-Xayah]

Can you please tell me how long should it take to generate data？Or publish the processed data？

[pschaldenbrand]

Hello, Could you please provide a little more information on the issue? It would be great to see what command you are running, so I can see if it is a problem with parameters. I'd also like more details on what you mean by "generate" and "processed" data.

Paintings generally take about 5-30 minutes to plan using my setup with an NVIDIA 4090 depending on number of brush strokes and optimization iterations. If yours is taking much longer than this, I think it's likely that you are not running this on the GPU. The code will take exceedingly long to run on CPU.

Best, Peter

[arturia-Xayah]

Hello, Could you please provide a little more information on the issue? It would be great to see what command you are running, so I can see if it is a problem with parameters. I'd also like more details on what you mean by "generate" and "processed" data.

Paintings generally take about 5-30 minutes to plan using my setup with an NVIDIA 4090 depending on number of brush strokes and optimization iterations. If yours is taking much longer than this, I think it's likely that you are not running this on the GPU. The code will take exceedingly long to run on CPU.

Best, Peter

Thanks for your help. I use CUDA_VISIBLE_DEVICES=0 bash create_copaint_data.sh to generate data with NVIDIA 3090 and I haven't changed any of the code in it . The command in create_copaint_data.sh is:

python3 create_copaint_data.py \
        --use_cache \
        --cache_dir ../src/caches/sharpie_short_strokes \
        --materials_json ../materials_ink.json \
        --lr_multiplier 2.0  \
        --n_iters 700  \
        --max_strokes_added 400  \
        --min_strokes_added 10  \
        --ink \
        --render_height 168 \
        --output_parent_dir train_data/ink

I checked that the value of the variable device(in create_copaint_data.py) is cuda. The running process is shown in the following figure.

Besides,it occupied about 8GB of CUDA memory. My python version is 3.11 with CUDA 12.2 .

[pschaldenbrand]

Thanks for that information, and sorry for the late reply (I was traveling).

The timings here seem accurate with my system. Each drawing takes quite a while to create. I left the script running for a day or two to generate the data for training CoFRIDA.

I could look into sharing the data I used to train it, but it is large and therefore difficult to host.