anything in a video

.gitattributes

@@ -36,3 +36,15 @@ rainbow.gif filter=lfs diff=lfs merge=lfs -text
 rainbow_.gif filter=lfs diff=lfs merge=lfs -text
 rainbow__.gif filter=lfs diff=lfs merge=lfs -text
 rainbow2.gif filter=lfs diff=lfs merge=lfs -text
+videos/jeep-moving.jpg filter=lfs diff=lfs merge=lfs -text
+videos/a_car_is_moving_on_the_road_40.mp4 filter=lfs diff=lfs merge=lfs -text
+videos/a_man_in_parkour_100.jpg filter=lfs diff=lfs merge=lfs -text
+videos/a_man_in_parkour_100.mp4 filter=lfs diff=lfs merge=lfs -text
+videos/child-riding_lego.jpg filter=lfs diff=lfs merge=lfs -text
+videos/child-riding_lego.mp4 filter=lfs diff=lfs merge=lfs -text
+videos/jeep-moving.mp4 filter=lfs diff=lfs merge=lfs -text
+videos/a_car_is_moving_on_the_road_40.jpg filter=lfs diff=lfs merge=lfs -text
+videos/a_man_is_surfing_3_30.jpg filter=lfs diff=lfs merge=lfs -text
+videos/a_man_is_surfing_3_30.mp4 filter=lfs diff=lfs merge=lfs -text
+videos/horse-running.jpg filter=lfs diff=lfs merge=lfs -text
+videos/horse-running.mp4 filter=lfs diff=lfs merge=lfs -text

app.py

@@ -53,6 +53,32 @@ def inference_mask1(prompt,
         res.append(np.uint8(np.array(Image.open(io.BytesIO(base64.b64decode(a[i]))))))
     return res
 
+
+
+def inference_mask_video(
+              prompt,
+              vid,
+              request: gr.Request,
+              ):
+
+
+    files = {
+        "pimage" : resizeImgIo(prompt["image"]),
+        "pmask" : resizeImgIo(prompt["mask"]),
+        "video" : open(vid, 'rb'),
+    }
+    r = requests.post("http://120.92.79.209/painter/runVideo", files = files)
+    '''
+    path = str(uuid.uuid4()) + "." + str(time.time())
+    fName = 'out.mp4'
+    file_out = "video/" + path + "." + fName
+    with open(file_out,"wb") as f:
+        f.write(r.content)
+    '''
+    a = json.loads(r.text)
+    return [np.uint8(np.array(Image.open(io.BytesIO(base64.b64decode(a["mask"]))))), a["url"]]
+
+
 def resizeImg(img):
     res, hres = 448, 448
     img = Image.fromarray(img).convert("RGB")
@@ -61,13 +87,13 @@ def resizeImg(img):
     img.save(temp, format="WEBP")
     return base64.b64encode(temp.getvalue()).decode('ascii')
 
-def inference_mask_cat(
-              prompt,
-              img,
-              img_,
-              ):
-    output_list = [img, img_]
-    return output_list
+def resizeImgIo(img):
+    res, hres = 448, 448
+    img = Image.fromarray(img).convert("RGB")
+    img = img.resize((res, hres))
+    temp = io.BytesIO()
+    img.save(temp, format="WEBP")
+    return io.BytesIO(temp.getvalue())
 
 
 # define app features and run
@@ -88,6 +114,15 @@ examples_sam = [
             ['./images/ydt_2.jpg', './images/ydt_1.jpg', './images/ydt_3.jpg'],
            ]
 
+examples_video = [
+            ['./videos/horse-running.jpg', './videos/horse-running.mp4'],
+            ['./videos/a_man_is_surfing_3_30.jpg', './videos/a_man_is_surfing_3_30.mp4'],
+    ['./videos/a_car_is_moving_on_the_road_40.jpg', './videos/a_car_is_moving_on_the_road_40.mp4'],
+['./videos/jeep-moving.jpg', './videos/jeep-moving.mp4'],
+['./videos/child-riding_lego.jpg', './videos/child-riding_lego.mp4'],
+]
+
+
 
 demo_mask = gr.Interface(fn=inference_mask1, 
                    inputs=[gr.ImageMask(brush_radius=8, label="prompt (提示图)"), gr.Image(label="img1 (测试图1)"), gr.Image(label="img2 (测试图2)")], 
@@ -134,6 +169,26 @@ demo_mask_sam = gr.Interface(fn=inference_mask1_sam,
                    allow_flagging="never",
                    )
 
+demo_mask_video = gr.Interface(fn=inference_mask_video,
+                   inputs=[gr.ImageMask(label="prompt (提示图)"), gr.Video(label="video (测试视频)").style(height=448, width=448)],
+                    outputs=[gr.Image(label="SAM output (mask)").style(height=256, width=256), gr.Video().style(height=448, width=448)],
+                    examples=examples_video,
+                    description="<p> \
+                    <strong>SegGPT+SAM: One touch for any segmentation in a video.</strong> <br>\
+                    Choose an example below &#128293; &#128293;  &#128293; <br>\
+                    Or, upload by yourself: <br>\
+                    1. Upload a video to be tested to 'video'. If failed, please check the codec, we recommend h.264 by default. <br>2. Upload a prompt image to 'prompt' and draw <strong>a point or line on the target</strong>.  <br>\
+<br> \
+💎 SAM segments the target with any point or scribble, then SegGPT segments the whole video. <br>\
+💎 Examples below were never trained and are randomly selected for testing in the wild. <br>\
+💎 Current UI interface only unleashes a small part of the capabilities of SegGPT, i.e., 1-shot case. <br> \
+                Note: we only take the first 16 frames for the demo.    \
+</p>",
+                   )
+
+
+
+
 title = "SegGPT: Segmenting Everything In Context<br> \
 <div align='center'> \
 <h2><a href='https://arxiv.org/abs/2304.03284' target='_blank' rel='noopener'>[paper]</a> \
@@ -144,7 +199,7 @@ title = "SegGPT: Segmenting Everything In Context<br> \
 </div> \
 "
 
-demo = gr.TabbedInterface([demo_mask_sam, demo_mask], ['SAM+SegGPT (一触百通)', 'General 1-shot'], title=title)
+demo = gr.TabbedInterface([demo_mask_sam, demo_mask_video, demo_mask], ['SAM+SegGPT (一触百通)', '🎬Anything in a Video', 'General 1-shot'], title=title)
 
 #demo.launch(share=True, auth=("baai", "vision"))
 demo.launch(enable_queue=False)

videos/a_car_is_moving_on_the_road_40.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e69783eed3294b0a76c147d46ce8705a46c21ae4122edc0fd9a2b57ee453954
+size 248227

videos/a_man_in_parkour_100.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dda65f31336de9f9ed607fe60443164a8529b98d949b7cb8e068a2245352e2e3
+size 1020054

videos/a_man_is_surfing_3_30.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73633b80752cefe0e68fb3bf6d1117d5bc1c094c4198c871beb7b59c6856f2f7
+size 301229

videos/child-riding_lego.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24a9ffd1a3a430851bb864f3fe1da4e84aed8d9694aea42d1f1578e7ef4818b4
+size 117189

videos/horse-running.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfd8b6c99776f291bf4c8787721387d8764c85b787741c665dee49dfb6442630
+size 383635

videos/jeep-moving.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92c2b38d7c52d8a19be7aa7f568d1d07b5fc433cbd369f45e028325230ad76ba
+size 150698