update README

Author: lmxyy

diffusion/README.md

@@ -51,6 +51,8 @@ Specifically,
 * `masks`  stores the difference masks between the original images and the edited images in `.npy` format.
 * `gt` stores the corresponding ground-truth images from [LSUN Church](https://github.com/fyu/lsun) dataset.
 
+If the dataset downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
+
 ## Get Started
 
 ### DDPM
@@ -73,7 +75,7 @@ We provide [![colab](https://colab.research.google.com/assets/colab-badge.svg)](
   --use_pretrained --save_dir results/sige_ddpm
   ```
 
-  The generated images will be stored in `results/vanilla_ddpm` and `results/sige_ddpm` correspondingly.
+  The generated images will be stored in `results/vanilla_ddpm` and `results/sige_ddpm` correspondingly. If the model downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
 
   **Notice**: For SIGE inference, current codebase only supports caching the activations of a single step. As the diffusion models require multiple denoising steps, for each step, we first forward use the original model to pre-compute the activations and then forward the SIGE model to get the results.
 
@@ -131,7 +133,7 @@ We provide [![colab](https://colab.research.google.com/assets/colab-badge.svg)](
   --use_pretrained --save_dir results/sige_pd
   ```
 
-  The generated images will be stored in `results/vanilla_pd` and `results/sige_pd` correspondingly.
+  The generated images will be stored in `results/vanilla_pd` and `results/sige_pd` correspondingly. If the model downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
 
   **Notice**: For SIGE inference, current codebase only supports caching the activations of a single step. As the diffusion models require multiple denoising steps, for each step, we first forward use the original model to pre-compute the activations and then forward the SIGE model to get the results.
 

diffusion_demo/README.md

@@ -45,7 +45,7 @@ SIGE achieves 2x less conversion time compared to original DDPM on M1 MacBook Pr
   python start.py --config_path configs/church_dpmsolver256-sige.yml
   ```
 
-By default, these commands will test results on GPU if GPU is available. You can also explicitly specify the device with `--device`. 
+By default, these commands will test results on GPU if GPU is available. You can also explicitly specify the device with `--device`. If the model downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
 
 
 ## Acknowledgement

gaugan/README.md

@@ -54,6 +54,8 @@ Specifically,
 * `synthetic_instances` and `synthetic_labels` store the synthetic instance labels and semantic labels, respectively.
 * `masks` stores the difference masks between the synthetic semantic maps and the corresponding ground-truth semantic maps in `.npy` format.
 
+If the dataset downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
+
 ## Get Started
 
 We provide [![colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/lmxyy/sige/blob/main/gaugan/gaugan.ipynb) to quickly test GauGAN model with SIGE.
@@ -88,6 +90,7 @@ We provide [![colab](https://colab.research.google.com/assets/colab-badge.svg)](
 
   * The results will be saved in the directory specified with `--save_dir`. For the generated results, `$id_synthetic.png` means using the synthetic map `$id` as the edited input and the corresponding ground-truth map as the original input. `$id_gt.png` means using the synthetic map `$id` as the original input and the corresponding ground-truth map as the edited input.
   * You could use `--dont_save_label` option to skip saving the input semantic maps in your results.
+  * If the model downloading is too slow for you, you can switch the download source from our website to Google Drive with `--download_tool gdown`.
 
 * Metric Measurement:
 

sige/nn/base.py

@@ -45,7 +45,6 @@ def load_runtime(self, function_name: str, runtime_dict: Dict = None):
                     os.environ["SIGE_METAL_LIB_PATH"] = os.path.abspath(
                         os.path.join(os.path.dirname(module.__file__), "..", "sige.metallib")
                     )
-                    print(os.environ["SIGE_METAL_LIB_PATH"])
             except (ModuleNotFoundError, AttributeError):
                 runtime_dict[device] = None
         return runtime_dict