[人工智能]【运行测试】使用encoder来generate图像

尝试利用generate.py 和训练好的encoder，来恢复图像
创建generate_encoder.py来做代码调整

project_name=StyleNeRF-main1
workdir=/host/home/joselyn/workspace/0419-course
outdir=$workdir/logs/$project_name


network=$workdir/stylenerf_pkl/ffhq_256.pkl
testIma_dir=$workdir/my_images/test_img.png
ffhq=$workdir/images256x256
encoder=$workdir/logs/StyleNeRF-main1/train_encoder1/checkpoints/network-snapshot-000240.pkl

python generate_encoder.py --outdir=$outdir/generate_encoder --trunc=0.7 --seeds=5 --network=$network --encoder=$encoder --render-program="rotation_camera"

debug时配置参数如下： --outdir=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main/generate_encoder --encoder=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main1/train_encoder1/checkpoints/network-snapshot-000240.pkl --trunc=0.7 --seeds=5 --network=/home/joselyn/workspace/0419-course/stylenerf_pkl/ffhq_256.pkl --render-program="rotation_camera:/home/joselyn/workspace/0419-course/my_images"

renderer.py line96 ws, _ = self.generator.encoder(kwargs['img'])
报错：RuntimeError: shape '[1, 17, 512]' is invalid for input of size 1470976

尝试调整输入原图像的分别率为256
修改debug参数配置：
--outdir=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main/generate_encoder --encoder=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main1/train_encoder1/checkpoints/network-snapshot-000240.pkl --trunc=0.7 --seeds=5 --network=/home/joselyn/workspace/0419-course/stylenerf_pkl/ffhq_256.pkl --render-program="rotation_camera:/home/joselyn/workspace/0419-course/my_images/aframes_256"
成功。
运行结果如下：

尝试换一种螺旋方式。
--outdir=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main/generate_encoder_rc3 --encoder=/home/joselyn/workspace/0419-course/logs/StyleNeRF-main1/train_encoder1/checkpoints/network-snapshot-000240.pkl --trunc=0.7 --seeds=5 --network=/home/joselyn/workspace/0419-course/stylenerf_pkl/ffhq_256.pkl --render-program="rotation_camera3:/home/joselyn/workspace/0419-course/my_images/aframes_256"

报错： File "/home/joselyn/workspace/0419-course/StyleNeRF-main1/renderer.py", line 145, in render_rotation_camera3 kwargs["latent_codes"] = gen.get_latent_codes(batch_size, tmp=self.sample_tmp, device=ws.device) AttributeError: 'tuple' object has no attribute 'device'
调整代码 renderer.py lin134
将ws = self.generator.encoder(kwargs['img'])['ws'] 改为ws, _ = self.generator.encoder(kwargs['img']) 从字面意思上看，两句话应该是一样的。
成功。

创作打卡挑战赛

赢取流量/现金/CSDN周边激励大奖