Need help in dealing with out of bounds

[seand0101]

Dear community members and @John6666

From the continuation of this problem, I have tried to update my id2label.json using this code to get the correct labels of every image and “pour it” to json file as in my dataset directory.

But the problem still persists and I hope some of the capable community members can help me look for the problem because I don’t know what to do anymore. I will keep looking for the solution but I guess more hands are faster than mine alone…

The error codes

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[17], line 1
----> 1 trainer.train()

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\trainer.py:2155, in Trainer.train(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
   2152 try:
   2153     # Disable progress bars when uploading models during checkpoints to avoid polluting stdout
   2154     hf_hub_utils.disable_progress_bars()
-> 2155     return inner_training_loop(
   2156         args=args,
   2157         resume_from_checkpoint=resume_from_checkpoint,
   2158         trial=trial,
   2159         ignore_keys_for_eval=ignore_keys_for_eval,
   2160     )
   2161 finally:
   2162     hf_hub_utils.enable_progress_bars()

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\trainer.py:2522, in Trainer._inner_training_loop(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)
   2516 context = (
   2517     functools.partial(self.accelerator.no_sync, model=model)
   2518     if i != len(batch_samples) - 1
   2519     else contextlib.nullcontext
   2520 )
   2521 with context():
-> 2522     tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
   2524 if (
   2525     args.logging_nan_inf_filter
   2526     and not is_torch_xla_available()
   2527     and (torch.isnan(tr_loss_step) or torch.isinf(tr_loss_step))
   2528 ):
   2529     # if loss is nan or inf simply add the average of previous logged losses
   2530     tr_loss = tr_loss + tr_loss / (1 + self.state.global_step - self._globalstep_last_logged)

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\trainer.py:3655, in Trainer.training_step(self, model, inputs, num_items_in_batch)
   3653         loss = self.compute_loss(model, inputs)
   3654     else:
-> 3655         loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
   3657 del inputs
   3658 if (
   3659     self.args.torch_empty_cache_steps is not None
   3660     and self.state.global_step % self.args.torch_empty_cache_steps == 0
   3661 ):

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\trainer.py:3709, in Trainer.compute_loss(self, model, inputs, return_outputs, num_items_in_batch)
   3707         loss_kwargs["num_items_in_batch"] = num_items_in_batch
   3708     inputs = {**inputs, **loss_kwargs}
-> 3709 outputs = model(**inputs)
   3710 # Save past state if it exists
   3711 # TODO: this needs to be fixed and made cleaner later.
   3712 if self.args.past_index >= 0:

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\modules\module.py:1736, in Module._wrapped_call_impl(self, *args, **kwargs)
   1734     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1735 else:
-> 1736     return self._call_impl(*args, **kwargs)

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\modules\module.py:1747, in Module._call_impl(self, *args, **kwargs)
   1742 # If we don't have any hooks, we want to skip the rest of the logic in
   1743 # this function, and just call forward.
   1744 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1745         or _global_backward_pre_hooks or _global_backward_hooks
   1746         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1747     return forward_call(*args, **kwargs)
   1749 result = None
   1750 called_always_called_hooks = set()

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\models\segformer\modeling_segformer.py:809, in SegformerForSemanticSegmentation.forward(self, pixel_values, labels, output_attentions, output_hidden_states, return_dict)
    807 if self.config.num_labels > 1:
    808     loss_fct = CrossEntropyLoss(ignore_index=self.config.semantic_loss_ignore_index)
--> 809     loss = loss_fct(upsampled_logits, labels)
    810 elif self.config.num_labels == 1:
    811     valid_mask = ((labels >= 0) & (labels != self.config.semantic_loss_ignore_index)).float()

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\modules\module.py:1736, in Module._wrapped_call_impl(self, *args, **kwargs)
   1734     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1735 else:
-> 1736     return self._call_impl(*args, **kwargs)

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\modules\module.py:1747, in Module._call_impl(self, *args, **kwargs)
   1742 # If we don't have any hooks, we want to skip the rest of the logic in
   1743 # this function, and just call forward.
   1744 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1745         or _global_backward_pre_hooks or _global_backward_hooks
   1746         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1747     return forward_call(*args, **kwargs)
   1749 result = None
   1750 called_always_called_hooks = set()

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\modules\loss.py:1293, in CrossEntropyLoss.forward(self, input, target)
   1292 def forward(self, input: Tensor, target: Tensor) -> Tensor:
-> 1293     return F.cross_entropy(
   1294         input,
   1295         target,
   1296         weight=self.weight,
   1297         ignore_index=self.ignore_index,
   1298         reduction=self.reduction,
   1299         label_smoothing=self.label_smoothing,
   1300     )

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\torch\nn\functional.py:3479, in cross_entropy(input, target, weight, size_average, ignore_index, reduce, reduction, label_smoothing)
   3477 if size_average is not None or reduce is not None:
   3478     reduction = _Reduction.legacy_get_string(size_average, reduce)
-> 3479 return torch._C._nn.cross_entropy_loss(
   3480     input,
   3481     target,
   3482     weight,
   3483     _Reduction.get_enum(reduction),
   3484     ignore_index,
   3485     label_smoothing,
   3486 )

IndexError: Target 225 is out of bounds.

The codes

model_checkpoint = "nvidia/segformer-b0-finetuned-ade-512-512"
from datasets import load_dataset

hf_dataset_identifier = "seand0101/segformer-b0-finetuned-ade-512-512-manggarai-watergate"

ds = load_dataset(hf_dataset_identifier)

ds

im_d = ds["train"]["label"][0].convert("L")
im_d

from huggingface_hub import hf_hub_download

import json

filename = "id2label.json"

id2label = json.load(

open(hf_hub_download(hf_dataset_identifier, filename, repo_type="dataset"), "r")

)

id2label = {int(k): v for k, v in id2label.items()}

label2id = {v: k for k, v in id2label.items()}

num_labels = len(id2label)

unique_labels = np.unique(id2label)

unique_labels

ds = ds.shuffle(seed=1)

ds = ds["train"].train_test_split(test_size=0.2)

train_ds = ds["train"]

test_ds = ds["test"]

from transformers import AutoFeatureExtractor

feature_extractor = AutoFeatureExtractor.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512")

from torchvision.transforms import ColorJitter
from transformers import SegformerFeatureExtractor

feature_extractor = SegformerFeatureExtractor()
jitter = ColorJitter(brightness=0.25, contrast=0.25, saturation=0.25, hue=0.1)

def train_transforms(example_batch):
    images = [jitter(x) for x in example_batch['pixel_values']]
    labels = [x.convert("L") for x in example_batch['label']]
    inputs = feature_extractor(images, labels)
    return inputs

def val_transforms(example_batch):
    images=[x for x in example_batch['pixel_values']]
    labels = [x.convert("L") for x in example_batch['label']]
    inputs = feature_extractor(images, labels)

    return inputs

# Set transforms
train_ds.set_transform(train_transforms)
test_ds.set_transform(val_transforms)

from transformers import SegformerForSemanticSegmentation

model = SegformerForSemanticSegmentation.from_pretrained(
    model_checkpoint,
    num_labels=num_labels,
    id2label=id2label,
    label2id=label2id,
    ignore_mismatched_sizes=True, # Will ensure the segmentation specific components are reinitialized. 
)

from transformers import TrainingArguments

epochs = 50
lr = 0.00006
batch_size = 2

hub_model_id = "nvidia/segformer-b0-finetuned-ade-512-512"

training_args = TrainingArguments(
    "segformer-b0-finetuned-ade20k-manggarai_rivergate",
    learning_rate=lr,
    num_train_epochs=epochs,
    per_device_train_batch_size=batch_size,
    per_device_eval_batch_size=batch_size,
    save_total_limit=3,
    eval_strategy="steps",
    save_strategy="steps",
    save_steps=20,
    eval_steps=20,
    logging_steps=1,
    eval_accumulation_steps=5,
    load_best_model_at_end=True,
    push_to_hub=True,
    hub_model_id=hub_model_id,
    hub_strategy="end",
)

import torch
from torch import nn
import evaluate

metric = evaluate.load("mean_iou")

def compute_metrics(eval_pred):
  with torch.no_grad():
    logits, labels = eval_pred
    logits_tensor = torch.from_numpy(logits)
    # scale the logits to the size of the label
    logits_tensor = nn.functional.interpolate(
        logits_tensor,
        size=labels.shape[-2:],
        mode="bilinear",
        align_corners=False,
    ).argmax(dim=1)

    pred_labels = logits_tensor.detach().cpu().numpy()
    # currently using _compute instead of compute
    # see this issue for more info: https://github.com/huggingface/evaluate/pull/328#issuecomment-1286866576
    metrics = metric._compute(
            predictions=pred_labels,
            references=labels,
            num_labels=len(id2label),
            ignore_index=0,
            reduce_labels=feature_extractor.reduce_labels,
        )
    
    # add per category metrics as individual key-value pairs
    per_category_accuracy = metrics.pop("per_category_accuracy").tolist()
    per_category_iou = metrics.pop("per_category_iou").tolist()

    metrics.update({f"accuracy_{id2label[i]}": v for i, v in enumerate(per_category_accuracy)})
    metrics.update({f"iou_{id2label[i]}": v for i, v in enumerate(per_category_iou)})
    
    return metrics

print(len(id2label))

from transformers import Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    tokenizer=feature_extractor,
    train_dataset=train_ds,
    eval_dataset=test_ds,
    compute_metrics=compute_metrics,
)

trainer.train()

I did it in jupyter notebook so sorry if the layout is bit weird. I copy everything on in case there’s something I see as a right thing to do but actually the center of the problem.

Thank you in advance-

Oh the code I use to extract the labels into id2label.json.

from transformers import SegformerForSemanticSegmentation, SegformerFeatureExtractor

import torch

from torchvision import transforms

from PIL import Image

# Load the pre-trained model and feature extractor

model = SegformerForSemanticSegmentation.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512")

feature_extractor = SegformerFeatureExtractor.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512")

# example for testing a new image

image_path = large_image_stack_512[0]

image = Image.open(image_path).convert("RGB")

image

# Prepare the image by applying transformations
inputs = feature_extractor(images=image, return_tensors="pt")

# Perform inference on the image
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits # Predicted logits for each pixel
   
# Manually define a mapping (example, you can get this from a dataset like ADE20k)
id2label = {
"0": "wall",
    "1": "building",
    "2": "sky",
    "3": "floor",
    "4": "tree",
    "5": "ceiling",
    "6": "road",
    "7": "bed ",
    "8": "windowpane",
    "9": "grass",
    "10": "cabinet",
    "11": "sidewalk",
    "12": "person",
    "13": "earth",
    "14": "door",
    "15": "table",
    "16": "mountain",
    "17": "plant",
    "18": "curtain",
    "19": "chair",
    "20": "car",
    "21": "water",
    "22": "painting",
    "23": "sofa",
    "24": "shelf",
    "25": "house",
    "26": "sea",
    "27": "mirror",
    "28": "rug",
    "29": "field",
    "30": "armchair",
    "31": "seat",
    "32": "fence",
    "33": "desk",
    "34": "rock",
    "35": "wardrobe",
    "36": "lamp",
    "37": "bathtub",
    "38": "railing",
    "39": "cushion",
    "40": "base",
    "41": "box",
    "42": "column",
    "43": "signboard",
    "44": "chest of drawers",
    "45": "counter",
    "46": "sand",
    "47": "sink",
    "48": "skyscraper",
    "49": "fireplace",
    "50": "refrigerator",
    "51": "grandstand",
    "52": "path",
    "53": "stairs",
    "54": "runway",
    "55": "case",
    "56": "pool table",
    "57": "pillow",
    "58": "screen door",
    "59": "stairway",
    "60": "river",
    "61": "bridge",
    "62": "bookcase",
    "63": "blind",
    "64": "coffee table",
    "65": "toilet",
    "66": "flower",
    "67": "book",
    "68": "hill",
    "69": "bench",
    "70": "countertop",
    "71": "stove",
    "72": "palm",
    "73": "kitchen island",
    "74": "computer",
    "75": "swivel chair",
    "76": "boat",
    "77": "bar",
    "78": "arcade machine",
    "79": "hovel",
    "80": "bus",
    "81": "towel",
    "82": "light",
    "83": "truck",
    "84": "tower",
    "85": "chandelier",
    "86": "awning",
    "87": "streetlight",
    "88": "booth",
    "89": "television receiver",
    "90": "airplane",
    "91": "dirt track",
    "92": "apparel",
    "93": "pole",
    "94": "land",
    "95": "bannister",
    "96": "escalator",
    "97": "ottoman",
    "98": "bottle",
    "99": "buffet",
    "100": "poster",
    "101": "stage",
    "102": "van",
    "103": "ship",
    "104": "fountain",
    "105": "conveyer belt",
    "106": "canopy",
    "107": "washer",
    "108": "plaything",
    "109": "swimming pool",
    "110": "stool",
    "111": "barrel",
    "112": "basket",
    "113": "waterfall",
    "114": "tent",
    "115": "bag",
    "116": "minibike",
    "117": "cradle",
    "118": "oven",
    "119": "ball",
    "120": "food",
    "121": "step",
    "122": "tank",
    "123": "trade name",
    "124": "microwave",
    "125": "pot",
    "126": "animal",
    "127": "bicycle",
    "128": "lake",
    "129": "dishwasher",
    "130": "screen",
    "131": "blanket",
    "132": "sculpture",
    "133": "hood",
    "134": "sconce",
    "135": "vase",
    "136": "traffic light",
    "137": "tray",
    "138": "ashcan",
    "139": "fan",
    "140": "pier",
    "141": "crt screen",
    "142": "plate",
    "143": "monitor",
    "144": "bulletin board",
    "145": "shower",
    "146": "radiator",
    "147": "glass",
    "148": "clock",
    "149": "flag"
}

import os

# get images mask
#Importing Dataset Pictures
ROOT_DIR = os.path.abspath("D:\\512\\512_adjust")
i = 1
large_image_stack_512= []
#print(i)
#directory = os.fsencode(ROOT_DIR)
print("Uploading Image from:", ROOT_DIR)
for file in os.listdir(ROOT_DIR):
    filename = os.fsdecode(file)
    if filename.endswith("jpg") or filename.endswith(".jpeg"):
        file = os.path.join(ROOT_DIR, filename)
        #print(filename)
        large_image_stack_512.append(file)
        #large_image_stack = np.array(filename)
        i+=1
        #print(i)
        continue
    else:
        continue

print("Image Address Retrived:", len(large_image_stack_512))

for_json = []
json_fill_flag = 0
content_i = 0
print(len(for_json))

f = open("forjson.txt", "w")

for im in large_image_stack_512:
    inputs = feature_extractor(images=Image.open(im).convert("RGB"), return_tensors="pt")

    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits # Predicted logits for each pixel

    predicted_mask = torch.argmax(logits, dim=1).squeeze().cpu().numpy()
    unique_class_indices = np.unique(predicted_mask)
    print("image_path", im)
    print(f"unique_class_indices: {unique_class_indices}")
    for class_idx in unique_class_indices:
        class_id = class_idx
        image_label = id2label.get(str(class_idx))
        #print(f"{class_idx}: {id2label.get(str(class_idx))}")
        new_f = tuple((class_id, image_label))
        if new_f not in for_json:
            for_json.append(new_f)

print(for_json)

The output of the List

[(0, 'wall'),
 (1, 'building'),
 (2, 'sky'),
 (4, 'tree'),
 (13, 'earth'),
 (21, 'water'),
 (32, 'fence'),
 (38, 'railing'),
 (61, 'bridge'),
 (103, 'ship'),
 (140, 'pier'),
 (16, 'mountain'),
 (122, 'tank'),
 (46, 'sand'),
 (26, 'sea'),
 (12, 'person'),
 (94, 'land'),
 (3, 'floor'),
 (53, 'stairs'),
 (109, 'swimming pool'),
 (60, 'river'),
 (34, 'rock'),
 (76, 'boat'),
 (6, 'road'),
 (20, 'car'),
 (83, 'truck')]

[seand0101]

Solved by putting these limiters on the code:

from torchvision.transforms import ColorJitter
from transformers import SegformerFeatureExtractor


jitter = ColorJitter(brightness=0.25, contrast=0.25, saturation=0.25, hue=0.1)

def train_transforms(example_batch):
    # images = [jitter(x) for x in example_batch['pixel_values']]
    # labels = [x.convert("L") for x in example_batch['label']]
    # inputs = feature_extractor(images, labels)
    # return inputs
    images = [jitter(x) for x in example_batch['pixel_values']]
    labels = [x.convert("L") for x in example_batch['label']]  # Convert label images to L mode (greyscale)
    labels = [np.array(label) for label in labels]  # Convert labels to numpy arrays for clamping
    labels = [np.clip(label, 0, num_labels - 1) for label in labels]  # Clamp labels to the valid range
    inputs = feature_extractor(images, labels, return_tensors="pt")
    return inputs

def val_transforms(example_batch):
    # images=[x for x in example_batch['pixel_values']]
    # labels = [x.convert("L") for x in example_batch['label']]
    # inputs = feature_extractor(images, labels)

    # return inputs
    images = [x for x in example_batch['pixel_values']]
    labels = [x.convert("L") for x in example_batch['label']]
    labels = [np.array(label) for label in labels]
    labels = [np.clip(label, 0, num_labels - 1) for label in labels]
    inputs = feature_extractor(images, labels, return_tensors="pt")
    return inputs

# Set transforms
train_ds.set_transform(train_transforms)
test_ds.set_transform(val_transforms)

And correcting this, I didn’t realize I redefining it again with empty feature extractor ()

from transformers import SegformerFeatureExtractor

feature_extractor = SegformerFeatureExtractor.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512")

I will update as to how it works after reading the docs, but so far I need to handle the training memories… what is a good epoch for a model and how do you decide that?

[John6666]

This works correctly, so the structure of the data set may be different from what HF expects. Returning a tensor for the Trainer function is correct.

#hf_dataset_identifier = "seand0101/segformer-b0-finetuned-ade-512-512-manggarai-watergate"
hf_dataset_identifier = "huggingface/semantic-segmentation-test-sample"

[seand0101]

Wait what do you mean by that, could you elaborate? Thank you in advance.

[John6666]

Oh, sorry, maybe I misunderstood. When I tested your code with the HF data set before your correction code came out, I was able to start training normally. So I thought there was a possibility that we could fix the data set in a simpler way.

[seand0101]

Really? So it’s not the code then? Okay let’s try to solve it on other way.

Btw I made my first model ever on my page, could you try it out? I received this error for some reason.

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
Cell In[6], line 15
     13 print(os.path.join(ROOT_DIR,im))
     14 # Semantic segmentation
---> 15 segmentation = markicob(jpeg_im)
     16 print("the length of current segmentation labels are: ", len(segmentation))
     17 while(mask_i < len(segmentation)):

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\pipelines\image_segmentation.py:144, in ImageSegmentationPipeline.__call__(self, inputs, **kwargs)
    142 if inputs is None:
    143     raise ValueError("Cannot call the image-classification pipeline without an inputs argument!")
--> 144 return super().__call__(inputs, **kwargs)

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\pipelines\base.py:1301, in Pipeline.__call__(self, inputs, num_workers, batch_size, *args, **kwargs)
   1293     return next(
   1294         iter(
   1295             self.get_iterator(
   (...)
   1298         )
   1299     )
   1300 else:
-> 1301     return self.run_single(inputs, preprocess_params, forward_params, postprocess_params)

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\pipelines\base.py:1309, in Pipeline.run_single(self, inputs, preprocess_params, forward_params, postprocess_params)
   1307 model_inputs = self.preprocess(inputs, **preprocess_params)
   1308 model_outputs = self.forward(model_inputs, **forward_params)
-> 1309 outputs = self.postprocess(model_outputs, **postprocess_params)
   1310 return outputs

File c:\Users\Lenovo\miniconda3\envs\pretrain-huggingface\Lib\site-packages\transformers\pipelines\image_segmentation.py:216, in ImageSegmentationPipeline.postprocess(self, model_outputs, subtask, threshold, mask_threshold, overlap_mask_area_threshold)
    214         mask = (segmentation == label) * 255
    215         mask = Image.fromarray(mask.astype(np.uint8), mode="L")
--> 216         label = self.model.config.id2label[label]
    217         annotation.append({"score": None, "label": label, "mask": mask})
    218 else:

KeyError: 5

How do you debug this?

[John6666]

maybe id2label[5] is missing?
I think it has to be a continuous sequence starting from 0.

As for the data set, if we compare these two, there must be an answer.

huggingface/semantic-segmentation-test-sample · Datasets at Hugging Face

seand0101/segformer-b0-finetuned-ade-512-512-manggarai-watergate · Datasets...

Edit:
The sample data set for HF is in grayscale, and the code is also for grayscale, so we may need to tweak the code for monochrome use.

DAGS - Daphne Koller's Research Group

[seand0101]

Okay so I’ll update the id2label again wait.

[seand0101]

changed the id2label.

Btw what I meant by try it on your side I kinda need to know your opinion on this model. As I wait I will compare both models as your recommendation

[John6666]

Oops. The punchline is that the mask data format required by SegFormer was special. If it’s a river or not, it’s either 0 or 1, so I think it’s possible to make it just by doing calculations.

Fine-Tune a Semantic Segmentation Model with a Custom Dataset

For semantic segmentation, we need a semantic bitmap that contains a category_id for each pixel. We’ll use the get_semantic_bitmap function from the Segments.ai SDK to convert the bitmaps to semantic bitmaps. To apply this function to all the rows in our dataset, we’ll use dataset.map.

[seand0101]

@John6666

The sample data set for HF is in grayscale, and the code is also for grayscale, so we may need to tweak the code for monochrome use.

Didn’t we just make them grayscale yesterday, I thought that was fixed already

Btw I did the inference with my model, it’s bad, do you think I need to do more epoch or need to just make the id2label just 1: “water” or 0: “not water”

[John6666]

need to just make the id2label just 1: “water” or 0: “not water”

Yes, because the fewer things you have to classify, the easier it is to classify them. Besides, the only thing in the actual training data is a river. Since it can only learn labels corresponding to the number of colors, it can only distinguish between two types when learning from black and white.

[seand0101]

I’m trying to do it with id2label only with 0 and 1, but I got this error with no clear documentation on how to do it. Do you know how to do it?

 'SegformerImageProcessor' object has no attribute 'reduce_labels'

So far I did it like this

from transformers import SegformerImageProcessor

feature_extractor = SegformerImageProcessor.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512", do_reduce_labels=True)

[John6666]

Maybe document is old…
https://github.com/huggingface/transformers/issues/25801

BTW, I think it will probably work even if there are only two items (0 and 1), but SegFormer may not recognize it properly unless it is a mask that has passed through the conversion function mentioned above. Even if it does work, there is a possibility that it will operate incorrectly. It seems that it is possible to reproduce it using only numpy operations, but I think it is better to use a function.

[seand0101]

I have read that github I cannot find heads or tails of the conversation, are they the maintainers?

Everything we do so far, I think, is doable with numpy and opencv . Before coming here I spent like 6 months trying to reverse engineer such code base but I got way too confused.

Huggingface transformers library felt better, but when there’s bug like this the layer we are now are already too high to debug the problem compared to those big codebases. Pluses and minuses I guess…

[John6666]

I’m also a relative newcomer to HF, github and Python, so I don’t know whether they’re maintainers or just passing through.
But I agree about libraries. Being familiar with the quirks of a library contributes to the results, rather than understanding the logic. Well, that’s just the way it is with existing resources.

[seand0101]

I don’t think I found the solution for this one, I will revert back to AutoFeatureExtractor

Wondering the result will be different if I don’t use the designated feature extractor