Update app.py

app.py

@@ -110,7 +110,7 @@ with gr.Blocks(css=css, theme=gr.themes.Base()) as demo:
                 [`Website`](https://4m.epfl.ch) | [`GitHub`](https://github.com/apple/ml-4m) <br>[`4M Paper (NeurIPS'23)`](https://arxiv.org/abs/2312.06647) | [`4M-21 Paper (arXiv'24)`](https://arxiv.org/abs/2406.09406)
 
                 This demo predicts all modalities from a given RGB input, using [{FM_MODEL_ID}](https://huggingface.co/{FM_MODEL_ID}), running on *{power_device}*.
-                For more generative examples, and to enable human pose visualizations, please see our [GitHub repo](https://github.com/apple/ml-4m).                
+                For more generative examples, please see our [GitHub repo](https://github.com/apple/ml-4m).                
                 """)
                 
                 img_path = gr.Image(label='RGB input image', type='filepath')
@@ -122,7 +122,7 @@ with gr.Blocks(css=css, theme=gr.themes.Base()) as demo:
                             ('CLIP-B/16', 'tok_clip@224'), ('DINOv2-B/14', 'tok_dinov2@224'), ('ImageBind-H/14', 'tok_imagebind@224'), 
                             ('Depth', 'tok_depth@224'), ('Surface normals', 'tok_normal@224'), ('Semantic segmentation', 'tok_semseg@224'), 
                             ('Canny edges', 'tok_canny_edge@224'), ('SAM edges', 'tok_sam_edge@224'), ('Caption', 'caption'), 
-                            ('Bounding boxes', 'det'), ('SAM instances', 'sam_instance'), ('Color palette', 'color_palette'), 
+                            ('Bounding boxes', 'det'), ('SAM instances (single pass*)', 'sam_instance'), ('Color palette', 'color_palette'), 
                             ('Metadata', 'metadata'),
                         ],
                         value=[
@@ -134,6 +134,21 @@ with gr.Blocks(css=css, theme=gr.themes.Base()) as demo:
                         label="Target modalities", 
                         info='Choose which modalities are predicted (in this order).'
                     )
+                    gr.Markdown(f"""
+                    **Information on modalities**:
+                    
+                    \* *SAM instances* in this demo are generated in a single pass and may look sparse. For sampling dense SAM instances, please see the convenience function 
+                    [`generate_sam_dense`](https://github.com/apple/ml-4m/blob/e11539965e45aa6731143d742c4493c46b4ef620/fourm/models/generate.py#L1230-L1273) 
+                    in `fourm.models.generate.GenerationSampler`, and our [4M-21 interactive notebook](https://github.com/apple/ml-4m/blob/main/notebooks/generation_4M-21.ipynb) for usage examples.
+                    
+                    \*\* While 4M-21 models are capable of predicting *4D human poses*, visualizing them requires the SMPL model which cannot be distributed.
+                    To visualize poses, please follow these steps:
+                
+                    1) Install via `pip install timm yacs smplx pyrender pyopengl==3.1.4`
+                    You may need to follow the pyrender install instructions: https://pyrender.readthedocs.io/en/latest/install/index.html
+                    2) Download SMPL data from https://smpl.is.tue.mpg.de/. See https://github.com/shubham-goel/4D-Humans/ for an example.
+                    3) Copy the required SMPL files (`smpl_mean_params.npz`, `SMPL_to_J19.pkl`, `smpl/SMPL_NEUTRAL.pkl`) to `fourm/utils/hmr2_utils/data` .
+                    """)
                     seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
                     randomize_seed = gr.Checkbox(label="Randomize seed", value=False)
                     top_p = gr.Slider(label="Top-p", minimum=0.0, maximum=1.0, step=0.01, value=0.8)