[Fixed] To skip images in the Save .* To Path nodes

When they are blocked

src/nodes/e_measure.py

@@ -7,7 +7,7 @@ def get_e_measure(
     pred: torch.Tensor,
     gt: torch.Tensor,
     num_thresholds: int = F_POINTS,
-    chunk_size: int = 16
+    chunk_size: int = -1
 ) -> Tuple[float, float, float, torch.Tensor, torch.Tensor]:
     """
     Calculates the E-measure scores using a memory-efficient chunking strategy.
@@ -38,6 +38,10 @@ def get_e_measure(
     thlist = torch.linspace(0, 1 - 1e-10, num_thresholds, device=pred.device)
     all_scores = []
 
+    if chunk_size == -1:
+        chunk_size = round(64 / (pred.shape[0] * pred.shape[1] / (1 << 20)))
+        chunk_size = min(max(1, chunk_size), num_thresholds)
+
     # Process thresholds in memory-efficient chunks
     for i in range(0, num_thresholds, chunk_size):
         # Get the current chunk of thresholds

src/nodes/helpers.py

@@ -147,13 +147,16 @@ def get_image_preview_info(file_name, where="input"):
     return {"filename": fname, "subfolder": dname, "type": where}
 
 
-def load_one_image(file_name, disp_name, embed_transparency):
+def load_one_image(file_name, disp_name, embed_transparency, known=None):
     if file_name is None:
         return (empty_image(b=1, c=3), empty_image(b=1), file_name)
     if not os.path.isabs(file_name):
         file_name = os.path.join(get_input_directory(), file_name)
     if not os.path.exists(file_name):
         raise ValueError(f"File '{file_name}' not found")
+    if known is not None and file_name in known:
+        logger.debug('Recycling '+file_name)
+        return (known[file_name][0], known[file_name][1], file_name)
 
     try:
         if has_load_image:
@@ -182,6 +185,8 @@ def load_one_image(file_name, disp_name, embed_transparency):
             # Concatenate image and mask into (b, h, w, 4)
             image_with_alpha = torch.cat([image, 1.0 - mask], dim=-1)
             result = (image_with_alpha, mask, file_name)
+        if known is not None:
+            known[file_name] = (result[0], result[1])
         return result
 
     except Exception as e:
@@ -191,13 +196,16 @@ def load_one_image(file_name, disp_name, embed_transparency):
                       "It may be corrupt or in an unsupported format.") from e
 
 
-def load_one_mask(file_name, disp_name, channel='red'):
+def load_one_mask(file_name, disp_name, channel='red', known=None):
     if file_name is None:
         return (empty_image(b=1), file_name)
     if not os.path.isabs(file_name):
         file_name = os.path.join(get_input_directory(), file_name)
     if not os.path.exists(file_name):
         raise ValueError(f"File '{file_name}' not found")
+    if known is not None and file_name in known:
+        logger.debug('Recycling '+file_name)
+        return (known[file_name], file_name)
 
     try:
         if has_load_image:
@@ -210,6 +218,8 @@ def load_one_mask(file_name, disp_name, channel='red'):
 
         # Call the method and return its result directly
         result = loader_instance.load_image(file_name, channel)
+        if known is not None:
+            known[file_name] = result[0]
         return (result[0], file_name)
 
     except Exception as e:
@@ -241,6 +251,8 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
 
     imgs = []
     masks = []
+    known_imgs = {}
+    known_masks = {}
     is_mask = channel is not None
     used_file_names = []
     all_preview_imgs = []
@@ -261,9 +273,9 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
 
                 if is_mask:
                     # A mask
-                    mask, file_name = load_one_mask(file_name, disp_name, channel)
+                    mask, file_name = load_one_mask(file_name, disp_name, channel, known=known_masks)
                 else:
-                    img, mask, file_name = load_one_image(file_name, disp_name, embed_transparency)
+                    img, mask, file_name = load_one_image(file_name, disp_name, embed_transparency, known=known_imgs)
 
                     max_w = max(max_w, img.shape[2])
                     max_h = max(max_h, img.shape[1])
@@ -279,7 +291,7 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
                     all_preview_imgs.append(get_image_preview_info(file_name))
 
             for j in range(len(imgs_batch)):
-                if channel is None:
+                if not is_mask:
                     img = imgs_batch[j]
                     H, W = img.shape[1:3]
                     if H != max_h or W != max_w:
@@ -292,7 +304,7 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
                     logger.debug(f"Upscaling mask to fit batch: {W}x{H} -> {max_mw}x{max_mh}")
                     masks_batch[j] = upscale_comfy(mask, max_mw, max_mh, "bicubic")
 
-            if channel is None:
+            if not is_mask:
                 imgs.append(torch.cat(imgs_batch))
             masks.append(torch.cat(masks_batch))
         else:
@@ -301,9 +313,9 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
 
             if is_mask:
                 # A mask
-                mask, file_name = load_one_mask(file_name, disp_names[i], channel)
+                mask, file_name = load_one_mask(file_name, disp_names[i], channel, known=known_masks)
             else:
-                img, mask, file_name = load_one_image(file_name, disp_names[i], embed_transparency)
+                img, mask, file_name = load_one_image(file_name, disp_names[i], embed_transparency, known=known_imgs)
                 imgs.append(img)
 
             masks.append(mask)
@@ -313,7 +325,8 @@ def load_images_wrapper(file_names, embed_transparency=False, disp_names=None, s
                 all_preview_imgs.append(get_image_preview_info(file_name))
 
     logger.debug(f"Loaded {len(masks)} batches:")
-    for n, i in enumerate(masks):
+    items = masks if is_mask else imgs
+    for n, i in enumerate(items):
         logger.debug(f"{n}) {i.shape}")
 
     result = (imgs, masks, used_file_names) if channel is None else (masks, used_file_names)
@@ -339,6 +352,8 @@ def save_image(images, filenames, prompt=None, extra_pnginfo=None, compress_leve
 
     all_preview_imgs = []
     for index, (image, filename) in enumerate(zip(imgs, filenames)):
+        if filename is None:
+            continue
         i = 255. * image.cpu().numpy()
         img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))
         metadata = PngInfo()

src/nodes/nodes_img.py

@@ -19,10 +19,11 @@ import re
 from seconohe.apply_mask import apply_mask
 from seconohe.color import color_to_rgb_float, color_to_rgb_uint8
 from seconohe.comfy_misc import (get_input_directory, get_output_directory, MAX_RESOLUTION, PromptServer, IO, ComfyNodeABC,
-                                 ExecutionBlocker, upscale_methods)
+                                 ExecutionBlocker, upscale_methods, ProgressBar, get_annotated_filepath)
 from seconohe.downloader import download_file
 from seconohe.foreground_estimation.affce import affce
 from seconohe.foreground_estimation.fmlfe import fmlfe, IMPL_PRIORITY
+from seconohe.misc import format_bytes
 from seconohe.tensor import batched_min_max_norm
 from seconohe.torch import get_default_comfy_device, get_canonical_device
 import torch
@@ -546,6 +547,12 @@ class ImageDataset(ComfyNodeABC):
                     "default": True,
                     "tooltip": "Block the execution when all images are processed"
                 }),
+                "invert": (IO.BOOLEAN, {
+                    "default": False,
+                    "tooltip": "Instead of listing the missing `destination` files\n"
+                               "we'll list the existing `destination` files.\n"
+                               "This can be used to collect generated information"
+                }),
             }
         }
 
@@ -565,7 +572,7 @@ class ImageDataset(ComfyNodeABC):
     @classmethod
     def IS_CHANGED(cls, source, pattern, destination, dest_ext, reference=None, sort_method="None",
                    image_load_cap=1, skip_first_images=0, select_every_nth=1, random_seed=1,
-                   block_when_finished=False):
+                   block_when_finished=False, invert=False):
         # We always indicate an evaluation is needed.
         # 1) We must check the directory to know it, so we don't have any advantage on doing it now.
         # 2) ComfyUI makes any check impossible, if source is connected to a node it will be None
@@ -573,7 +580,7 @@ class ImageDataset(ComfyNodeABC):
 
     def execute(self, source, pattern, destination, dest_ext, reference=None, sort_method="None",
                 image_load_cap=1, skip_first_images=0, select_every_nth=1, random_seed=1,
-                block_when_finished=False):
+                block_when_finished=False, invert=False):
         logger.debug(f"ImageDataset.source = {source}")
         logger.debug(f"ImageDataset.destination = {destination}")
         # Here self isn't really needed, our state is the filesystem
@@ -643,7 +650,7 @@ class ImageDataset(ComfyNodeABC):
             dest_path = dest_dir / dest_filename
 
             # Skip if the result file already exists
-            if dest_path.exists():
+            if dest_path.exists() ^ invert:
                 continue
 
             # Find the reference file (case-insensitive and extension-agnostic)
@@ -794,8 +801,60 @@ class SaliencyEvaluationMetrics(ComfyNodeABC):
     UNIQUE_NAME = "SET_SaliencyEvaluationMetrics"
     DISPLAY_NAME = "Saliency Evaluation Metrics"
 
+    def read_single(self, path):
+        if not path.is_file():
+            return None
+        with open(path, "rt") as f:
+            f.readline()
+            return float(f.readline())
+
+    def read_f(self, path):
+        if not path.is_file():
+            return None, None, None
+        f4 = torch.Tensor(size=(4, F_POINTS), device="cpu")
+        f1 = f4.movedim(1, 0)  # Easier here, not for computation
+        with open(path, "rt") as f:
+            reader = csv.reader(f)
+            next(reader)  # header
+            for c in range(F_POINTS):
+                f1[c] = torch.Tensor(tuple(float(v) for v in next(reader)))
+            next(reader)
+            row = next(reader)
+            assert row[0] == "Max", row
+            f_max = float(row[1])
+            row = next(reader)
+            assert row[0] == "Adaptive", row
+            f_adp = float(row[1])
+            return f_max, f4, f_adp
+
+    def read_e(self, path):
+        if not path.is_file():
+            return None, None, None, None, None
+        with open(path, "rt") as f:
+            reader = csv.reader(f)
+            next(reader)  # header
+            all_e = []
+            thres = []
+            for c in range(F_POINTS):
+                row = next(reader)
+                thres.append(float(row[0]))
+                all_e.append(float(row[1]))
+            thres = torch.Tensor(thres)
+            all_e = torch.Tensor(all_e)
+            next(reader)
+            row = next(reader)
+            assert row[0] == "Mean"
+            e_mean = float(row[1])
+            row = next(reader)
+            assert row[0] == "Max"
+            e_max = float(row[1])
+            row = next(reader)
+            assert row[0] == "Adaptive"
+            e_adp = float(row[1])
+            return e_mean, e_max, e_adp, all_e, thres
+
     def evaluate(self, prediction: torch.Tensor, ground_truth: torch.Tensor, unique_id,
-                 img_name, normalize, result_save: bool = False, mae_enable: bool = True, mae_save: bool = False,
+                 img_name, normalize: bool = False, result_save: bool = False, mae_enable: bool = True, mae_save: bool = False,
                  max_f_mes_enable: bool = True, max_f_mes_save: bool = False, s_mes_enable: bool = True,
                  s_mes_save: bool = False, e_mes_enable: bool = True, e_mes_save: bool = False,
                  wf_mes_enable: bool = True, wf_mes_save: bool = True):
@@ -835,6 +894,7 @@ class SaliencyEvaluationMetrics(ComfyNodeABC):
         if names_len != imgs_len:
             raise ValueError(f"Got {imgs_len} images and {names_len} names, they must match")
 
+        prog_bar = ProgressBar(pred_len)
         for index_img in range(pred_len):
             # Ensure tensors are on the same device
             inputs_are_copies = get_canonical_device(prediction[index_img].device) != device
@@ -866,75 +926,105 @@ class SaliencyEvaluationMetrics(ComfyNodeABC):
 
                 # 1. Mean Absolute Error (MAE)
                 if mae_enable:
-                    mae = torch.mean(torch.abs(pred_i - gt_i)).item()
-                    logger.debug(f"MAE: {mae}")
+                    mae_path = Path(imgp.parent, imgp.stem+"_mae.csv")
+                    mae = self.read_single(mae_path)
+                    if mae is None:
+                        mae = torch.mean(torch.abs(pred_i - gt_i)).item()
+                        logger.debug(f"MAE: {mae}")
+                        if mae_save:
+                            with open(mae_path, "wt") as f:
+                                f.write(f"MAE\n{mae}")
+                    else:
+                        logger.debug(f"MAE already computed: {mae}")
                     mae_total += mae
                     res['mae'] = mae
-                    if mae_save:
-                        with open(Path(imgp.parent, imgp.stem+"_mae.csv"), "wt") as f:
-                            f.write(f"MAE\n{mae}")
 
                 # --- Metrics requiring binary ground truth ---
-                if max_f_mes_enable or s_mes_enable or e_mes_enable or wf_mes_enable:
-                    gt_binary = (gt_i >= 0.5).float()
+                gt_binary = None
 
                 # 2. Max F-measure
                 if max_f_mes_enable:
-                    f_max, all_f, f_adp = get_f_measure(pred_i, gt_binary)
+                    f_path = Path(imgp.parent, imgp.stem+"_F.csv")
+                    f_max, all_f, f_adp = self.read_f(f_path)
+                    if f_max is None:
+                        gt_binary = (gt_i >= 0.5).float() if gt_binary is None else gt_binary
+                        f_max, all_f, f_adp = get_f_measure(pred_i, gt_binary)
+                        logger.debug(f"Fβmax: {f_max}")
+                        if max_f_mes_save:
+                            all_fp = all_f.movedim(1, 0)
+                            with open(f_path, "wt") as f:
+                                f.write("Threshold, F-measure, Precision, Recall\n")
+                                for fn in all_fp:
+                                    f.write(f"{fn[0]}, {fn[1]}, {fn[2]}, {fn[3]}\n")
+                                f.write(f"\nMax, {f_max}\n")
+                                f.write(f"Adaptive, {f_adp}\n")
+                    else:
+                        logger.debug(f"Fβmax already computed: {f_max}")
                     f_measure_max_total += f_max
-                    logger.debug(f"Fβmax: {f_max}")
                     res['max_f_mes'] = f_max
                     res['adp_f_mes'] = f_adp
                     res['f'] = all_f[1]
                     res['fp'] = all_f[2]
                     res['fr'] = all_f[3]
-                    if max_f_mes_save:
-                        with open(Path(imgp.parent, imgp.stem+"_F.csv"), "wt") as f:
-                            f.write("Threshold, F-measure\n")
-                            for fn in all_f:
-                                f.write(f"{fn[0]}, {fn[1]}\n")
-                            f.write(f"\nMax, {f_max}\n")
 
                 # 3. S-measure
                 if s_mes_enable:
-                    s_measure = get_s_measure(pred_i, gt_binary)
+                    s_path = Path(imgp.parent, imgp.stem+"_S.csv")
+                    s_measure = self.read_single(s_path)
+                    if s_measure is None:
+                        gt_binary = (gt_i >= 0.5).float() if gt_binary is None else gt_binary
+                        s_measure = get_s_measure(pred_i, gt_binary)
+                        logger.debug(f"Sα: {s_measure}")
+                        if s_mes_save:
+                            with open(s_path, "wt") as f:
+                                f.write(f"S-measure\n{s_measure}")
+                    else:
+                        logger.debug(f"Sα already computed: {s_measure}")
                     s_measure_total += s_measure
-                    logger.debug(f"Sα: {s_measure}")
                     res['s_mes'] = s_measure
-                    if s_mes_save:
-                        with open(Path(imgp.parent, imgp.stem+"_S.csv"), "wt") as f:
-                            f.write(f"S-measure\n{s_measure}")
 
                 # 4. E-measure
                 if e_mes_enable:
-                    e_mean, e_max, e_adp, all_e, thres = get_e_measure(pred_i, gt_binary)
+                    e_path = Path(imgp.parent, imgp.stem+"_E.csv")
+                    e_mean, e_max, e_adp, all_e, thres = self.read_e(e_path)
+                    if e_mean is None:
+                        gt_binary = (gt_i >= 0.5).float() if gt_binary is None else gt_binary
+                        e_mean, e_max, e_adp, all_e, thres = get_e_measure(pred_i, gt_binary)
+                        logger.debug(f"Eϕ: {e_mean} {e_max} {e_adp}")
+                        if e_mes_save:
+                            with open(e_path, "wt") as f:
+                                f.write("Threshold, E-measure\n")
+                                for index, en in enumerate(all_e):
+                                    f.write(f"{thres[index]}, {en}\n")
+                                f.write("\n")
+                                f.write(f"Mean, {e_mean}\n")
+                                f.write(f"Max, {e_max}\n")
+                                f.write(f"Adaptive, {e_adp}\n")
+                    else:
+                        logger.debug(f"Eϕ already computed: {e_mean} {e_max} {e_adp}")
                     e_measure_total += e_mean
                     e_measure_max_total += e_max
                     e_measure_adp_total += e_adp
-                    logger.debug(f"Eϕ: {e_mean} {e_max} {e_adp}")
                     res['e_mes'] = e_mean
                     res['max_e_mes'] = e_max
                     res['adp_e_mes'] = e_adp
                     res['e'] = all_e
-                    if e_mes_save:
-                        with open(Path(imgp.parent, imgp.stem+"_E.csv"), "wt") as f:
-                            f.write("Threshold, E-measure\n")
-                            for index, en in enumerate(all_e):
-                                f.write(f"{thres[index]}, {en}\n")
-                            f.write("\n")
-                            f.write(f"Mean, {e_mean}\n")
-                            f.write(f"Max, {e_max}\n")
-                            f.write(f"Adaptive, {e_adp}\n")
 
                 # 5. Weighted F-measure
                 if wf_mes_enable:
-                    wf = get_weighted_f_measure(pred_i, gt_binary)
+                    wf_path = Path(imgp.parent, imgp.stem+"_wF.csv")
+                    wf = self.read_single(wf_path)
+                    if wf is None:
+                        gt_binary = (gt_i >= 0.5).float() if gt_binary is None else gt_binary
+                        wf = get_weighted_f_measure(pred_i, gt_binary)
+                        logger.debug(f"Fβw: {wf}")
+                        if wf_mes_save:
+                            with open(wf_path, "wt") as f:
+                                f.write(f"Weighted F-measure\n{wf}")
+                    else:
+                        logger.debug(f"Fβw already computed: {wf}")
                     weighted_f_total += wf
-                    logger.debug(f"Fβw: {wf}")
                     res['wf_mes'] = wf
-                    if wf_mes_save:
-                        with open(Path(imgp.parent, imgp.stem+"_wF.csv"), "wt") as f:
-                            f.write(f"Weighted F-measure\n{wf}")
 
                 if result_save and res:
                     with open(Path(imgp.parent, imgp.stem+".csv"), "wt") as f:
@@ -943,6 +1033,7 @@ class SaliencyEvaluationMetrics(ComfyNodeABC):
                         f.write(','.join(expand_data_row(res, keys))+"\n")
 
                 all.append(res)
+                prog_bar.update(1)
 
         # Average metrics over the batch/es
         mae_avg = mae_total / gt_len
@@ -1115,8 +1206,8 @@ class ConsolidateMetrics(ComfyNodeABC):
             raise ValueError(f"Got {n_metrics} metrics and {n_destination} destinations. They must be multiples.")
         if n_res_destination > 1:
             raise ValueError(f"Only one resume destination currently supported, got {n_res_destination}")
-        if n_res_destination and n_metric_ids != n_metrics:
-            raise ValueError(f"Got {n_metrics} metrics and {n_metric_ids} metric IDs. They must match.")
+        if n_res_destination and n_metric_ids != n_destination:
+            raise ValueError(f"Got {n_destination} destinations and {n_metric_ids} metric IDs. They must match.")
         slice_size = n_metrics // n_destination
 
         res = []
@@ -1126,7 +1217,7 @@ class ConsolidateMetrics(ComfyNodeABC):
             end = start + slice_size
             data, total = self.consolidate(metrics[start:end], img_name[start:end], d)
             res.append(data)
-            if n_res_destination:
+            if n_res_destination and total:
                 totals[metric_ids[c]] = total
 
         if totals:
@@ -2521,6 +2612,29 @@ class ImagePad(ComfyNodeABC):
         return (out_image, out_masks)
 
 
+class ChooseUpscaleMethod(ComfyNodeABC):
+    """
+    Used to send the same upscale option to other nodes.
+    """
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "upscale_method": UPSCALE_OPT,
+            },
+        }
+
+    RETURN_TYPES = (IO.ANY,)
+    RETURN_NAMES = ("upscale_method",)
+    FUNCTION = "execute"
+    CATEGORY = BASE_CATEGORY + "/" + MANIPULATION_CATEGORY
+    UNIQUE_NAME = "SET_ChooseUpscaleMethod"
+    DISPLAY_NAME = "Choose Upscale Method"
+
+    def execute(self, upscale_method):
+        return (upscale_method, )
+
+
 # Adapted from KJNodes, credits to Kijai (ImageResizeKJv2)
 # Differences:
 # - The color is an string that support various formats
@@ -3027,6 +3141,9 @@ class ImageWithTextLabel(ComfyNodeABC):
 
 
 class CartesianProduct(ComfyNodeABC):
+    """
+    Produces a list with all the combinations from the items in list_A and list_B.
+    """
     @classmethod
     def INPUT_TYPES(s):
         return {
@@ -3071,3 +3188,110 @@ class CartesianProduct(ComfyNodeABC):
                     res_B.append(b)
 
         return (res_A, res_B)
+
+
+class ImageMemoryEstimator:
+    """
+    Estimates the ammount of RAM needed to load the indicated images.
+    """
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "file_name": (IO.STRING, {
+                    "forceInput": True,
+                    "tooltip": "One or more image names. ComfyUI input dir is the base"}),
+                "is_image": (IO.BOOLEAN, {
+                    "default": True,
+                    "tooltip": "Compute three channels. Enable it and `is_mask` for RGBA"}),
+                "is_mask": (IO.BOOLEAN, {
+                    "default": False,
+                    "tooltip": "Compute one channel. Can be combined with `is_image`"}),
+            },
+            "hidden": {
+                "unique_id": "UNIQUE_ID",
+            },
+        }
+
+    INPUT_IS_LIST = True
+    RETURN_TYPES = ("INT", "STRING", "STRING")
+    RETURN_NAMES = ("total_bytes", "human_readable", "file_name")
+    OUTPUT_IS_LIST = (False, False, True)
+    FUNCTION = "execute"
+    CATEGORY = BASE_CATEGORY + "/" + VALIDATION
+    UNIQUE_NAME = "SET_ImageMemoryEstimator"
+    DISPLAY_NAME = "Image Memory Estimator"
+
+    def execute(self, file_name, is_image, is_mask, unique_id):
+        total_pixels = 0
+        min_area = float('inf')
+        max_area = 0
+        min_dim = (0, 0)
+        max_dim = (0, 0)
+        valid_count = 0
+
+        # Bytes per pixel for ComfyUI Tensors (Float32 = 4 bytes, RGB = 3 channels)
+        channels = 3 if is_image[0] else 0
+        channels += 1 if is_mask[0] else 0
+        if channels == 0:
+            logger.warning("Enable `is_image` and/or `is_mask`")
+        BYTES_PER_PIXEL = 4 * channels
+        logger.debug(f"BYTES_PER_PIXEL {BYTES_PER_PIXEL}")
+
+        known = set()
+        for fname in file_name:
+            if fname is None:
+                continue
+            # Try to resolve path (handles both absolute paths and ComfyUI input folder files)
+            file_path = get_annotated_filepath(fname)
+
+            if not file_path or not os.path.exists(file_path):
+                logger.warning(f"Could not find file: {fname}")
+                continue
+
+            if file_path in known:
+                continue
+
+            try:
+                # PIL.Image.open is lazy; it reads metadata without loading pixel data
+                with Image.open(file_path) as img:
+                    w, h = img.size
+                    area = w * h
+                    total_pixels += area
+
+                    if area < min_area:
+                        min_area = area
+                        min_dim = (w, h)
+
+                    if area > max_area:
+                        max_area = area
+                        max_dim = (w, h)
+
+                    valid_count += 1
+                    known.add(file_path)
+            except Exception as e:
+                logger.error(f"Failed to read metadata for {fname}: {e}")
+
+        total_bytes = total_pixels * BYTES_PER_PIXEL
+        human_readable = format_bytes(total_bytes)
+
+        # --- Information ---
+        logger.info("--- Image Memory Estimation ---")
+        msg = f"Number of images: {valid_count}"
+        logger.info(msg)
+        if valid_count > 0:
+            m = f"Minimum Dimensions: {min_dim[0]}x{min_dim[1]}"
+            logger.info(m)
+            msg += "<br>" + m
+            m = f"Maximum Dimensions: {max_dim[0]}x{max_dim[1]}"
+            logger.info(m)
+            msg += "<br>" + m
+            m = f"Total Raw Tensor Size: {human_readable} ({total_bytes} bytes)"
+            logger.info(m)
+            msg += "<br>" + m
+        else:
+            logger.info("No valid images processed.")
+        logger.info("-------------------------------")
+        send_progress_text(unique_id[0], msg)
+
+        return (total_bytes, human_readable, file_name)