Use HF Hub for model downloads

app.py

@@ -2,6 +2,7 @@ import os
 os.system("pip install ./MultiScaleDeformableAttention-1.0-py3-none-any.whl")
 
 import gradio as gr
+from huggingface_hub import hf_hub_download
 import numpy as np
 import numpy as np
 import torch
@@ -61,8 +62,8 @@ def show_box(box, ax):
     ax.add_patch(plt.Rectangle((x0, y0), w, h, edgecolor='green', facecolor=(0,0,0,0), lw=2))
 
 
-
-sam_checkpoint = r"~/.cache/huggingface/hub/andzhang01/segment_anything/sam_vit_l_0b3195.pth"
+sam_checkpoint = hf_hub_download(repo_id="andzhang01/segment-anything", filename="sam_vit_l_0b3195.pth")
+# sam_checkpoint = r"~/.cache/huggingface/hub/models--andzhang01--segment-anything/sam_vit_l_0b3195.pth"
 model_type = "vit_l"
 
 sam = sam_model_registry[model_type](checkpoint=sam_checkpoint, device=device)
@@ -180,8 +181,11 @@ def predict_one(net, image, mask, box, transforms, hypar, device):
 
 hypar = {} # paramters for inferencing
 
-hypar["model_path"] ="~/.cache/huggingface/hub/jwlarocque/DIS-SAM"
-hypar["restore_model"] = "DIS-SAM-checkpoint.pth"
+dis_model_path = hf_hub_download(repo_id="jwlarocque/DIS-SAM", filename="DIS-SAM-checkpoint.pth")
+# hypar["model_path"] ="~/.cache/huggingface/hub/jwlarocque/DIS-SAM"
+hypar["model_path"] = os.path.split(dis_model_path)[0]
+# hypar["restore_model"] = "DIS-SAM-checkpoint.pth"
+hypar["restore_model"] = os.path.split(dis_model_path)[1]
 hypar["model_digit"] = "full"
 hypar["input_size"] = [1024, 1024]
 hypar["model"] = ISNetDIS(in_ch=5)