search

jzhangbs / Vis-MVSNet

Visibility-aware Multi-view Stereo Network
MIT License
235 stars 27 forks source link

Caught NanError in replica 0 on device 0. #8

Open NoOneUST opened 3 years ago

NoOneUST commented 3 years ago

Hello. I first run the command python sh/bld.py local, the sh/dir.json file is set as:

{
    "local": {
        "bld_dir": "dataset_low_res",
        "dtu_dir": "",
        "tnt_dir": "tankandtemples",
        "tnt_training_dir": "",
        "save_dir": "save",
        "batch_size": 4,
        "num_workers": 4,
        "train_environ": "",
        "val_environ": ""
    }
}

Then I set the sh/bld_val.py file as:

import os
import json
import argparse

parser = argparse.ArgumentParser()
parser.add_argument('machine', type=str)
args = parser.parse_args()

with open('sh/dir.json') as f:
    d = json.load(f)
d = d[args.machine]

for m in ['temp']:
    for ns in range(3, 3+1):
        cmd = f"""{d['val_environ']}
            python val.py
            --data_root {d['bld_dir']}
            --dataset_name blended
            --model_name model_cas
            --num_src {ns}
            --max_d 128
            --interval_scale 1
            --cas_depth_num 32,16,8
            --cas_interv_scale 4,2,1
            --resize 768,576
            --crop 768,576
            --mode soft
            --load_path {d['save_dir']}/{m}
        """
        cmd = ' '.join(cmd.strip().split())
        print(cmd)
        os.system(cmd)

Then, I run the command python sh/bld_val.py local and get:

(four cards condition)
(base) bash-4.2$ python sh/bld_val.py local
python val.py --data_root dataset_low_res --dataset_name blended --model_name model_cas --num_src 3 --max_d 128 --interval_scale 1 --cas_depth_num 32,16,8 --cas_interv_scale 4,2,1 --resize 768,576 --crop 768,576 --mode soft --load_path save/temp
Number of samples: 915
Number of model parameters: 1162696
load save/temp/-1
  0%|                                                                                                                                                                        | 0/915 [00:23<?, ?it/s]
Traceback (most recent call last):
  File "val.py", line 99, in <module>
    outputs, refined_depth, prob_maps = model(sample, cas_depth_num, cas_interv_scale, mode=args.mode)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 161, in forward
    outputs = self.parallel_apply(replicas, inputs, kwargs)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 171, in parallel_apply
    return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 86, in parallel_apply
    output.reraise()
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/_utils.py", line 428, in reraise
    raise self.exc_type(msg)
utils.utils.NanError: Caught NanError in replica 0 on device 0.
Original Traceback (most recent call last):
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/parallel/parallel_apply.py", line 61, in _worker
    output = module(*input, **kwargs)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 419, in forward
    est_depth_1, prob_map_1, pair_results_1 = self.stage1([ref_feat_1, ref_cam, srcs_feat_1, srcs_cam], depth_num=depth_nums[0], upsample=False, mem=mem, mode=mode, depth_start_override=None, depth_interval_override=depth_interval*interval_scales[0], s_scale=8)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 317, in forward
    warped_src = self.build_cost_volume(ref_feat, ref_cam, src_feat, src_cam, depth_num, depth_start, depth_interval, s_scale, d_scale)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 179, in build_cost_volume
    warped_src_nd_c_h_w = homography_warping(src_nd_c_h_w, Hs.view(-1, *Hs.size()[2:]))  # n*d chw
  File "/export/data/lwangcg/Vis-MVSNet/core/homography.py", line 103, in homography_warping
    warped = interpolate(input, warped_coord)
  File "/export/data/lwangcg/Vis-MVSNet/core/homography.py", line 93, in interpolate
    raise NanError
utils.utils.NanError
(single card condition)
(base) bash-4.2$ CUDA_VISIBLE_DEVICES=0 python sh/bld_val.py local
python val.py --data_root dataset_low_res --dataset_name blended --model_name model_cas --num_src 3 --max_d 128 --interval_scale 1 --cas_depth_num 32,16,8 --cas_interv_scale 4,2,1 --resize 768,576 --crop 768,576 --mode soft --load_path save/temp
Number of samples: 915
Number of model parameters: 1162696
load save/temp/-1
  0%|                                                                                                                                                                        | 0/915 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "val.py", line 99, in <module>
    outputs, refined_depth, prob_maps = model(sample, cas_depth_num, cas_interv_scale, mode=args.mode)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/parallel/data_parallel.py", line 159, in forward
    return self.module(*inputs[0], **kwargs[0])
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 419, in forward
    est_depth_1, prob_map_1, pair_results_1 = self.stage1([ref_feat_1, ref_cam, srcs_feat_1, srcs_cam], depth_num=depth_nums[0], upsample=False, mem=mem, mode=mode, depth_start_override=None, depth_interval_override=depth_interval*interval_scales[0], s_scale=8)
  File "/export/data/lwangcg/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 317, in forward
    warped_src = self.build_cost_volume(ref_feat, ref_cam, src_feat, src_cam, depth_num, depth_start, depth_interval, s_scale, d_scale)
  File "/export/data/lwangcg/Vis-MVSNet/core/model_cas.py", line 179, in build_cost_volume
    warped_src_nd_c_h_w = homography_warping(src_nd_c_h_w, Hs.view(-1, *Hs.size()[2:]))  # n*d chw
  File "/export/data/lwangcg/Vis-MVSNet/core/homography.py", line 103, in homography_warping
    warped = interpolate(input, warped_coord)
  File "/export/data/lwangcg/Vis-MVSNet/core/homography.py", line 93, in interpolate
    raise NanError
utils.utils.NanError

Could you please give me some help? Thank you!

jzhangbs commented 3 years ago

yes we may encounter nan when training with blendedmvs. but it should have been caught in the training loop