init project

app.py

@@ -448,16 +448,16 @@ def get_reconstructed_scene(outdir, pe3r, device, silent, filelist, schedule, ni
     images = Images(filelist=filelist, device=device)
     
     # try:
-    # cog_seg_maps, rev_cog_seg_maps, cog_feats = get_cog_feats(images, pe3r, device)
-    # imgs = load_images(images, rev_cog_seg_maps, size=512, verbose=not silent)
-    # except Exception as e:
-    rev_cog_seg_maps = []
-    for tmp_img in images.np_images:
-        rev_seg_map = -np.ones(tmp_img.shape[:2], dtype=np.int64)
-        rev_cog_seg_maps.append(rev_seg_map)
-    cog_seg_maps = rev_cog_seg_maps
-    cog_feats = torch.zeros((1, 1024))
+    cog_seg_maps, rev_cog_seg_maps, cog_feats = get_cog_feats(images, pe3r, device)
     imgs = load_images(images, rev_cog_seg_maps, size=512, verbose=not silent)
+    # except Exception as e:
+        # rev_cog_seg_maps = []
+        # for tmp_img in images.np_images:
+        #     rev_seg_map = -np.ones(tmp_img.shape[:2], dtype=np.int64)
+        #     rev_cog_seg_maps.append(rev_seg_map)
+        # cog_seg_maps = rev_cog_seg_maps
+        # cog_feats = torch.zeros((1, 1024))
+        # imgs = load_images(images, rev_cog_seg_maps, size=512, verbose=not silent)
 
     if len(imgs) == 1:
         imgs = [imgs[0], copy.deepcopy(imgs[0])]
@@ -469,56 +469,56 @@ def get_reconstructed_scene(outdir, pe3r, device, silent, filelist, schedule, ni
         scenegraph_type = scenegraph_type + "-" + str(refid)
 
     pairs = make_pairs(imgs, scene_graph=scenegraph_type, prefilter=None, symmetrize=True)
-    output = inference(pairs, pe3r.mast3r, device, batch_size=1, verbose=not silent)
-    mode = GlobalAlignerMode.PointCloudOptimizer if len(imgs) > 2 else GlobalAlignerMode.PairViewer
-    scene_1 = global_aligner(output, cog_seg_maps, rev_cog_seg_maps, cog_feats, device=device, mode=mode, verbose=not silent)
-    lr = 0.01
-    # if mode == GlobalAlignerMode.PointCloudOptimizer:
-    loss = scene_1.compute_global_alignment(tune_flg=True, init='mst', niter=niter, schedule=schedule, lr=lr)
-
-    try:
-        ImgNorm = tvf.Compose([tvf.ToTensor(), tvf.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
-        for i in range(len(imgs)):
-            # print(imgs[i]['img'].shape, scene.imgs[i].shape, ImgNorm(scene.imgs[i])[None])
-            imgs[i]['img'] = ImgNorm(scene_1.imgs[i])[None]
-        pairs = make_pairs(imgs, scene_graph=scenegraph_type, prefilter=None, symmetrize=True)
-        output = inference(pairs, pe3r.mast3r, device, batch_size=1, verbose=not silent)
-        mode = GlobalAlignerMode.PointCloudOptimizer if len(imgs) > 2 else GlobalAlignerMode.PairViewer
-        scene = global_aligner(output, cog_seg_maps, rev_cog_seg_maps, cog_feats, device=device, mode=mode, verbose=not silent)
-        ori_imgs = scene.ori_imgs
-        lr = 0.01
-        # if mode == GlobalAlignerMode.PointCloudOptimizer:
-        loss = scene.compute_global_alignment(tune_flg=False, init='mst', niter=niter, schedule=schedule, lr=lr)
-    except Exception as e:
-        scene = scene_1
-        scene.imgs = ori_imgs
-        scene.ori_imgs = ori_imgs
-        print(e)
-
+    # output = inference(pairs, pe3r.mast3r, device, batch_size=1, verbose=not silent)
+    # mode = GlobalAlignerMode.PointCloudOptimizer if len(imgs) > 2 else GlobalAlignerMode.PairViewer
+    # scene_1 = global_aligner(output, cog_seg_maps, rev_cog_seg_maps, cog_feats, device=device, mode=mode, verbose=not silent)
+    # lr = 0.01
+    # # if mode == GlobalAlignerMode.PointCloudOptimizer:
+    # loss = scene_1.compute_global_alignment(tune_flg=True, init='mst', niter=niter, schedule=schedule, lr=lr)
 
-    outfile = get_3D_model_from_scene(outdir, silent, scene, min_conf_thr, as_pointcloud, mask_sky,
-                                      clean_depth, transparent_cams, cam_size)
-
-    # also return rgb, depth and confidence imgs
-    # depth is normalized with the max value for all images
-    # we apply the jet colormap on the confidence maps
-    rgbimg = scene.imgs
-    depths = to_numpy(scene.get_depthmaps())
-    confs = to_numpy([c for c in scene.im_conf])
-    # confs = to_numpy([c for c in scene.conf_2])
-    cmap = pl.get_cmap('jet')
-    depths_max = max([d.max() for d in depths])
-    depths = [d / depths_max for d in depths]
-    confs_max = max([d.max() for d in confs])
-    confs = [cmap(d / confs_max) for d in confs]
-
-    imgs = []
-    for i in range(len(rgbimg)):
-        imgs.append(rgbimg[i])
-        imgs.append(rgb(depths[i]))
-        imgs.append(rgb(confs[i]))
-
-    return scene, outfile, imgs
+    # try:
+    #     ImgNorm = tvf.Compose([tvf.ToTensor(), tvf.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
+    #     for i in range(len(imgs)):
+    #         # print(imgs[i]['img'].shape, scene.imgs[i].shape, ImgNorm(scene.imgs[i])[None])
+    #         imgs[i]['img'] = ImgNorm(scene_1.imgs[i])[None]
+    #     pairs = make_pairs(imgs, scene_graph=scenegraph_type, prefilter=None, symmetrize=True)
+    #     output = inference(pairs, pe3r.mast3r, device, batch_size=1, verbose=not silent)
+    #     mode = GlobalAlignerMode.PointCloudOptimizer if len(imgs) > 2 else GlobalAlignerMode.PairViewer
+    #     scene = global_aligner(output, cog_seg_maps, rev_cog_seg_maps, cog_feats, device=device, mode=mode, verbose=not silent)
+    #     ori_imgs = scene.ori_imgs
+    #     lr = 0.01
+    #     # if mode == GlobalAlignerMode.PointCloudOptimizer:
+    #     loss = scene.compute_global_alignment(tune_flg=False, init='mst', niter=niter, schedule=schedule, lr=lr)
+    # except Exception as e:
+    #     scene = scene_1
+    #     scene.imgs = ori_imgs
+    #     scene.ori_imgs = ori_imgs
+    #     print(e)
+
+
+    # outfile = get_3D_model_from_scene(outdir, silent, scene, min_conf_thr, as_pointcloud, mask_sky,
+    #                                   clean_depth, transparent_cams, cam_size)
+
+    # # also return rgb, depth and confidence imgs
+    # # depth is normalized with the max value for all images
+    # # we apply the jet colormap on the confidence maps
+    # rgbimg = scene.imgs
+    # depths = to_numpy(scene.get_depthmaps())
+    # confs = to_numpy([c for c in scene.im_conf])
+    # # confs = to_numpy([c for c in scene.conf_2])
+    # cmap = pl.get_cmap('jet')
+    # depths_max = max([d.max() for d in depths])
+    # depths = [d / depths_max for d in depths]
+    # confs_max = max([d.max() for d in confs])
+    # confs = [cmap(d / confs_max) for d in confs]
+
+    # imgs = []
+    # for i in range(len(rgbimg)):
+    #     imgs.append(rgbimg[i])
+    #     imgs.append(rgb(depths[i]))
+    #     imgs.append(rgb(confs[i]))
+
+    # return scene, outfile, imgs
 
 @spaces.GPU(duration=180)
 def get_3D_object_from_scene(outdir, pe3r, silent, device, text, threshold, scene, min_conf_thr, as_pointcloud,