Merge branch 'main' of https://huggingface.co/datasets/AstroCompress/GBI-16-4D into main

utils/eval_baselines.py

@@ -89,30 +89,30 @@ def main(dim):
 
     for path in tqdm(file_paths):
         with fits.open(path) as hdul:
-            if dim == '2d':
+            if dim == '2d': # compress the first timestep frame, R wavelength band (index 2)
                 arr = hdul[0].data[0][2]
                 arrs = [arr]
-            elif dim == '2d-top':
+            elif dim == '2d-top': # same as 2d, but only top 8 bits. This is to compare with similarly preprocessed neural approaches.
                 arr = hdul[0].data[0][2]
                 arr = split_uint16_to_uint8(arr)[0]
                 arrs = [arr]
-            elif dim == '2d-bottom':
+            elif dim == '2d-bottom': # same as 2d, but only bottom 8 bits. This is to compare with similarly preprocessed neural approaches.
                 arr = hdul[0].data[0][2]
                 arr = split_uint16_to_uint8(arr)[1]
                 arrs = [arr]
-            elif dim == '3dt' and len(hdul[0].data) > 2:
+            elif dim == '3dt' and len(hdul[0].data) > 2: # 3D tensor with first 3 timestep frames of wavelength band index 2
                 arr = hdul[0].data[0:3][2]
                 arrs = [arr]
-            elif dim == '3dw' and len(hdul[0].data[0]) > 2:
+            elif dim == '3dw' and len(hdul[0].data[0]) > 2: # 3D tensor with first timestep frame on wavelength bands of indices 1,2,3 (G, R, I bands)
                 arr = hdul[0].data[0][0:3]
                 arrs = [arr]
-            elif dim == '3dt_reshape' and len(hdul[0].data) > 2:
+            elif dim == '3dt_reshape' and len(hdul[0].data) > 2: # Same as 3dt but reshape into 2D array, for compatibility with JPEG-LS and RICE
                 arr = hdul[0].data[0:3][2].reshape((800, -1))
                 arrs = [arr]
-            elif dim == '3dw_reshape' and len(hdul[0].data[0]) > 2:
+            elif dim == '3dw_reshape' and len(hdul[0].data[0]) > 2: # Same as 3dw but reshape into 2D array, for compatibility with JPEG-LS and RICE
                 arr = hdul[0].data[0][0:3].reshape((800, -1))
                 arrs = [arr]
-            elif dim == 'tw':
+            elif dim == 'tw': # Iterate through all possible arrays where the x,y spatial location is fixed, and the remaining 2D array consists of ALL timesteps, ALL wavelengths. 
                 init_arr = hdul[0].data
                 def arrs_gen():
                     for i in range(init_arr.shape[-2]):
@@ -153,7 +153,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "dimension", 
         choices=['2d', '2d-top', '2d-bottom', '3dt', '3dw', 'tw', '3dt_reshape', '3dw_reshape'],
-        help="Specify whether the data is 2d, 3dt (3d time dimension), or 3dw (3d wavelength dimension)."
+        help="Specify whether the data is 2d, 3dt (3d time dimension), 3dw (3d wavelength dimension), 2d-top (only top 8 bits), 2d-bottom (only bottom 8 bits), tw (only a single x,y spatial location but all timesteps and wavelengths), 3dt_reshape or 3dw_reshape for the 2D flattened 3D evals, for use on JPEG-LS or RICE."
     )
     args = parser.parse_args()
     dim = args.dimension.lower()

utils/sdss_downloading.txt

@@ -0,0 +1,2 @@
+Please visit this repo:
+https://github.com/profjsb/astrocompress/tree/main