search

neuralchen / SimSwap

An arbitrary face-swapping framework on images and videos with one single trained model!
Other
4.55k stars 895 forks source link

how to use the trained 512 model ? #251

Open doctorcui opened 2 years ago

Fibonacci134 commented 2 years ago

Please refrain from spamming issues, reread the Updated preparation guide and if you have a specific problem, im sure one of us can help you. Thanks 🙏

doctorcui commented 2 years ago

ok, thank you i have a problem now， i follow the guide to train 512 model first i prepare 6W pics and i run python scripts/crop_align_vggface2_FFHQalign.py --input_dir $DATAPATH$/VGGface2/train --output_dir_ffhqalign $ALIGN_OUTDIR$ --mode ffhq --crop_size 256 and python scripts/inference_gfpgan_forvggface2.py --input_path $ALIGN_OUTDIR$ --batchSize 8 --save_dir $HQ_OUTDIR$

to make my 512*512 dataset

then i run python train.py --name simswap512_test --batchSize 16 --gpu_ids 0 --dataset /path/to/VGGFace2HQ --Gdeep True to train my model

i use python test_video_swapsingle.py --crop_size 512 --use_mask --name simswap512 --Arc_path arcface_model/arcface_checkpoint.tar --pic_a_path ./demo_file/Iron_man.jpg --video_path ./demo_file/multi_people_1080p.mp4 --output_path ./output/multi_test_swapsingle.mp4 --temp_path ./temp_results to test but finallly error comes error appears []Pretrained network G has fewer layers; The following are not initialized: ['down0', 'first_layer', 'last_layer', 'up0'] I don't know where I went wrong I really look forward to your help

doctorcui commented 2 years ago

Finally I found the problem I have been following simswap continuously, which led me to try training for the first time I'm using the April 20th version train.py # Last Modified: Wednesday, 20th April 2022 6:33:30 pm but trian.py updated # Last Modified: Friday, 22nd April 2022 10:49:26 am remove line 182,183,184 if opt.train_simswap: src_image1 = F.interpolate(src_image1,size=(256,256), mode='bicubic') src_image2 = F.interpolate(src_image2,size=(256,256), mode='bicubic')

this might be my problem I hope you pay attention to the training version

Fibonacci134 commented 2 years ago

Hey, Im glad you found the issue! Good luck, will definitely be paying attention to see how your guy's model turns out. Remember to not undertrain but also, dont over train 🙂

netrunner-exe commented 2 years ago

ok, thank you i have a problem now， i follow the guide to train 512 model first i prepare 6W pics and i run python scripts/crop_align_vggface2_FFHQalign.py --input_dir $DATAPATH$/VGGface2/train --output_dir_ffhqalign $ALIGN_OUTDIR$ --mode ffhq --crop_size 256 and python scripts/inference_gfpgan_forvggface2.py --input_path $ALIGN_OUTDIR$ --batchSize 8 --save_dir $HQ_OUTDIR$

to make my 512*512 dataset

then i run python train.py --name simswap512_test --batchSize 16 --gpu_ids 0 --dataset /path/to/VGGFace2HQ --Gdeep True to train my model

i use python test_video_swapsingle.py --crop_size 512 --use_mask --name simswap512 --Arc_path arcface_model/arcface_checkpoint.tar --pic_a_path ./demo_file/Iron_man.jpg --video_path ./demo_file/multi_people_1080p.mp4 --output_path ./output/multi_test_swapsingle.mp4 --temp_path ./temp_results to test but finallly error comes error appears []Pretrained network G has fewer layers; The following are not initialized: ['down0', 'first_layer', 'last_layer', 'up0'] I don't know where I went wrong I really look forward to your help

If I'm not mistaken, the dataset published in the VGGFace2-HQ repository is already aligned, croped and processed with GFPGAN to 512