Add files using upload-large-folder tool

.gitattributes

@@ -33,17 +33,3 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
-demo_images/output.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/compare_current_ddim_50.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/scheduler_ablation/euler.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/output_custom_metadata.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/full_ddim_zeros.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/full_ddim_none.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/full_ddim_custom.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/full_pndm_custom.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/full_euler_custom.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/ablation/fp16_metadata_none.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/ablation/fp16_metadata_zeros.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/ablation/fp32_metadata_none.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/scheduler_ablation/ddim.jpeg filter=lfs diff=lfs merge=lfs -text
-demo_images/ablation/fp32_metadata_zeros.jpeg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -10,20 +10,16 @@ tags:
 widget:
   - prompt: satellite image of farmland with roads
     output:
-      url: demo_images/output.jpeg
+      url: demo_images/readme_text2img.jpeg
 ---
 
-> [!WARNING] we do not have a full checkpoint conversion validation, if you encounter pipeline loading failure and unsidered output, please contact me via bili_sakura@zju.edu.cn
+> [!NOTE]
+> If you encounter pipeline loading failure or unexpected output, please contact bili_sakura@zju.edu.cn.
 
 # DiffusionSat Custom Pipelines
 
 Custom community pipelines for loading DiffusionSat checkpoints directly with `diffusers.DiffusionPipeline.from_pretrained()`.
 
-> [!WARNING]
-> **Known checkpoint issue (current release):** generation can degrade to near-total noise for many prompts/settings.
-> This appears to be a checkpoint compatibility/conversion issue, not expected model behavior.
-> We are working on a fix and will update this model card when a corrected checkpoint is available.
-
 See [Diffusers Community Pipeline Documentation](https://huggingface.co/docs/diffusers/using-diffusers/custom_pipeline_overview).
 
 ## Model Index
@@ -68,47 +64,8 @@ pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
 image = pipe(
     "satellite image of farmland",
     metadata=None,  # Optional
+    height=512,
+    width=512,
     num_inference_steps=50,
 ).images[0]
 ```
-
-### 2. ControlNet Pipeline
-
-Use `pipeline_diffusionsat_controlnet.py` for ControlNet generation.
-
-```python
-import torch
-from diffusers import DiffusionPipeline, ControlNetModel
-from diffusers.utils import load_image
-
-# 1. Load ControlNet
-controlnet = ControlNetModel.from_pretrained(
-    "path/to/ckpt/diffusionsat/controlnet",
-    torch_dtype=torch.float16
-)
-
-# 2. Load Pipeline with ControlNet
-pipe = DiffusionPipeline.from_pretrained(
-    "path/to/ckpt/diffusionsat",
-    controlnet=controlnet,
-    custom_pipeline="./pipeline_diffusionsat_controlnet.py", # Path to this file
-    torch_dtype=torch.float16,
-    trust_remote_code=True,
-)
-pipe = pipe.to("cuda")
-
-# 3. Prepare Control Image
-control_image = load_image("path/to/conditioning_image.png")
-
-# 4. Generate
-# metadata: Target image metadata (optional)
-# cond_metadata: Conditioning image metadata (optional)
-
-image = pipe(
-    "satellite image of farmland",
-    image=control_image,
-    metadata=None,       
-    cond_metadata=None,
-    num_inference_steps=30,
-).images[0]
-```

scheduler/scheduler_config.json

@@ -6,7 +6,7 @@
   "beta_start": 0.00085,
   "clip_sample": false,
   "num_train_timesteps": 1000,
-  "prediction_type": "epsilon",
+  "prediction_type": "v_prediction",
   "set_alpha_to_one": false,
   "skip_prk_steps": true,
   "steps_offset": 1,