diff --git a/dvc/dependency/__init__.py b/dvc/dependency/__init__.py
index 91bd96c232..803deccfbe 100644
--- a/dvc/dependency/__init__.py
+++ b/dvc/dependency/__init__.py
@@ -49,9 +49,15 @@
 del SCHEMA[BaseOutput.PARAM_METRIC]
 SCHEMA.update(RepoDependency.REPO_SCHEMA)
 SCHEMA.update(ParamsDependency.PARAM_SCHEMA)
+SCHEMA.update({BaseOutput.PARAM_FILTER: str})
 
 
 def _get(stage, p, info):
+    if isinstance(p, dict):
+        p = list(p.items())
+        assert len(p) == 1
+        p, extra_info = p[0]  # PARAM_FILTER
+        info.update(extra_info)
     parsed = urlparse(p) if p else None
     if parsed and parsed.scheme == "remote":
         tree = get_cloud_tree(stage.repo, name=parsed.netloc)
diff --git a/dvc/output/base.py b/dvc/output/base.py
index f17e3df9e5..d5f303c090 100644
--- a/dvc/output/base.py
+++ b/dvc/output/base.py
@@ -52,6 +52,7 @@ class BaseOutput:
 
     PARAM_PATH = "path"
     PARAM_CACHE = "cache"
+    PARAM_FILTER = "cmd"
     PARAM_METRIC = "metric"
     PARAM_METRIC_TYPE = "type"
     PARAM_METRIC_XPATH = "xpath"
@@ -176,6 +177,10 @@ def checksum(self):
     def checksum(self, checksum):
         self.info[self.tree.PARAM_CHECKSUM] = checksum
 
+    @property
+    def filter_cmd(self):
+        return self.info.get(self.PARAM_FILTER)
+
     def get_checksum(self):
         return self.tree.get_hash(self.path_info)
 
@@ -188,7 +193,7 @@ def exists(self):
         return self.tree.exists(self.path_info)
 
     def save_info(self):
-        return self.tree.save_info(self.path_info)
+        return self.tree.save_info(self.path_info, cmd=self.filter_cmd)
 
     def changed_checksum(self):
         return self.checksum != self.get_checksum()
@@ -313,6 +318,8 @@ def dumpd(self):
         if self.persist:
             ret[self.PARAM_PERSIST] = self.persist
 
+        if self.filter_cmd:
+            ret[self.PARAM_FILTER] = self.filter_cmd
         return ret
 
     def verify_metric(self):
diff --git a/dvc/repo/tree.py b/dvc/repo/tree.py
index 8011d0318c..1f30e5024a 100644
--- a/dvc/repo/tree.py
+++ b/dvc/repo/tree.py
@@ -226,7 +226,8 @@ def isdvc(self, path, **kwargs):
     def isexec(self, path):  # pylint: disable=unused-argument
         return False
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         outs = self._find_outs(path_info, strict=False)
         if len(outs) != 1:
             raise OutputNotFoundError
@@ -404,7 +405,7 @@ def walk_files(self, top, **kwargs):  # pylint: disable=arguments-differ
             for fname in files:
                 yield PathInfo(root) / fname
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
         """Return file checksum for specified path.
 
         If path_info is a DVC out, the pre-computed checksum for the file
@@ -418,7 +419,7 @@ def get_file_hash(self, path_info):
                 return self.dvctree.get_file_hash(path_info)
             except OutputNotFoundError:
                 pass
-        return file_md5(path_info, self)[0]
+        return file_md5(path_info, self, cmd=cmd)[0]
 
     def copytree(self, top, dest):
         top = PathInfo(top)
diff --git a/dvc/schema.py b/dvc/schema.py
index 0948f1f613..3005e1c375 100644
--- a/dvc/schema.py
+++ b/dvc/schema.py
@@ -17,10 +17,12 @@
     StageParams.PARAM_ALWAYS_CHANGED: bool,
 }
 
-DATA_SCHEMA = {**CHECKSUMS_SCHEMA, Required("path"): str}
+DATA_SCHEMA = {**CHECKSUMS_SCHEMA, Required(BaseOutput.PARAM_PATH): str}
 LOCK_FILE_STAGE_SCHEMA = {
     Required(StageParams.PARAM_CMD): str,
-    StageParams.PARAM_DEPS: [DATA_SCHEMA],
+    StageParams.PARAM_DEPS: [
+        {**DATA_SCHEMA, Optional(BaseOutput.PARAM_FILTER): str}
+    ],
     StageParams.PARAM_PARAMS: {str: {str: object}},
     StageParams.PARAM_OUTS: [DATA_SCHEMA],
 }
@@ -51,7 +53,9 @@
     str: {
         StageParams.PARAM_CMD: str,
         Optional(StageParams.PARAM_WDIR): str,
-        Optional(StageParams.PARAM_DEPS): [str],
+        Optional(StageParams.PARAM_DEPS): [
+            Any(str, {str: {BaseOutput.PARAM_FILTER: str}})
+        ],
         Optional(StageParams.PARAM_PARAMS): [
             Any(str, PARAM_PSTAGE_NON_DEFAULT_SCHEMA)
         ],
diff --git a/dvc/tree/azure.py b/dvc/tree/azure.py
index 19f9ecdcb9..1829cff04b 100644
--- a/dvc/tree/azure.py
+++ b/dvc/tree/azure.py
@@ -139,7 +139,8 @@ def remove(self, path_info):
         logger.debug(f"Removing {path_info}")
         self.blob_service.delete_blob(path_info.bucket, path_info.path)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         return self.get_etag(path_info)
 
     def _upload(
diff --git a/dvc/tree/base.py b/dvc/tree/base.py
index 1860bafab6..828a379411 100644
--- a/dvc/tree/base.py
+++ b/dvc/tree/base.py
@@ -63,6 +63,7 @@ class BaseTree:
     CACHE_MODE = None
     SHARED_MODE_MAP = {None: (None, None), "group": (None, None)}
     PARAM_CHECKSUM = None
+    PARAM_FILTER = None
 
     state = StateNoop()
 
@@ -236,7 +237,7 @@ def is_dir_hash(cls, hash_):
             return False
         return hash_.endswith(cls.CHECKSUM_DIR_SUFFIX)
 
-    def get_hash(self, path_info, **kwargs):
+    def get_hash(self, path_info, cmd=None, **kwargs):
         assert path_info and (
             isinstance(path_info, str) or path_info.scheme == self.scheme
         )
@@ -265,14 +266,14 @@ def get_hash(self, path_info, **kwargs):
         if self.isdir(path_info):
             hash_ = self.get_dir_hash(path_info, **kwargs)
         else:
-            hash_ = self.get_file_hash(path_info)
+            hash_ = self.get_file_hash(path_info, cmd=cmd)
 
         if hash_ and self.exists(path_info):
             self.state.save(path_info, hash_)
 
         return hash_
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
         raise NotImplementedError
 
     def get_dir_hash(self, path_info, **kwargs):
@@ -293,8 +294,11 @@ def path_to_hash(self, path):
 
         return "".join(parts)
 
-    def save_info(self, path_info, **kwargs):
-        return {self.PARAM_CHECKSUM: self.get_hash(path_info, **kwargs)}
+    def save_info(self, path_info, cmd=None, **kwargs):
+        ret = {self.PARAM_CHECKSUM: self.get_hash(path_info, **kwargs)}
+        if cmd:
+            ret[self.PARAM_FILTER] = cmd
+        return ret
 
     def _calculate_hashes(self, file_infos):
         file_infos = list(file_infos)
diff --git a/dvc/tree/gdrive.py b/dvc/tree/gdrive.py
index abf48d17ab..799663adad 100644
--- a/dvc/tree/gdrive.py
+++ b/dvc/tree/gdrive.py
@@ -573,7 +573,7 @@ def remove(self, path_info):
         item_id = self._get_item_id(path_info)
         self.gdrive_delete_file(item_id)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
         raise NotImplementedError
 
     def _upload(self, from_file, to_info, name=None, no_progress_bar=False):
diff --git a/dvc/tree/gs.py b/dvc/tree/gs.py
index 767154de35..df4473abab 100644
--- a/dvc/tree/gs.py
+++ b/dvc/tree/gs.py
@@ -182,7 +182,8 @@ def copy(self, from_info, to_info):
         to_bucket = self.gs.bucket(to_info.bucket)
         from_bucket.copy_blob(blob, to_bucket, new_name=to_info.path)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         import base64
         import codecs
 
diff --git a/dvc/tree/hdfs.py b/dvc/tree/hdfs.py
index cdc9a2c339..d064718eb8 100644
--- a/dvc/tree/hdfs.py
+++ b/dvc/tree/hdfs.py
@@ -161,7 +161,8 @@ def _group(regex, s, gname):
         assert match is not None
         return match.group(gname)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         # NOTE: pyarrow doesn't support checksum, so we need to use hadoop
         regex = r".*\t.*\t(?P<checksum>.*)"
         stdout = self.hadoop_fs(
diff --git a/dvc/tree/http.py b/dvc/tree/http.py
index abed95ca4f..b62042fe3c 100644
--- a/dvc/tree/http.py
+++ b/dvc/tree/http.py
@@ -125,7 +125,8 @@ def request(self, method, url, **kwargs):
     def exists(self, path_info, use_dvcignore=True):
         return bool(self.request("HEAD", path_info.url))
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         url = path_info.url
         headers = self.request("HEAD", url).headers
         etag = headers.get("ETag") or headers.get("Content-MD5")
diff --git a/dvc/tree/local.py b/dvc/tree/local.py
index bfce3345f8..f686e5f437 100644
--- a/dvc/tree/local.py
+++ b/dvc/tree/local.py
@@ -29,6 +29,7 @@ class LocalTree(BaseTree):
     scheme = Schemes.LOCAL
     PATH_CLS = PathInfo
     PARAM_CHECKSUM = "md5"
+    PARAM_FILTER = "cmd"
     PARAM_PATH = "path"
     TRAVERSE_PREFIX_LEN = 2
     UNPACKED_DIR_SUFFIX = ".unpacked"
@@ -297,8 +298,8 @@ def is_protected(self, path_info):
 
         return stat.S_IMODE(mode) == self.CACHE_MODE
 
-    def get_file_hash(self, path_info):
-        return file_md5(path_info)[0]
+    def get_file_hash(self, path_info, cmd=None):
+        return file_md5(path_info, cmd=cmd)[0]
 
     @staticmethod
     def getsize(path_info):
diff --git a/dvc/tree/s3.py b/dvc/tree/s3.py
index 44368d7461..7a4a45c7c5 100644
--- a/dvc/tree/s3.py
+++ b/dvc/tree/s3.py
@@ -317,7 +317,8 @@ def _copy(cls, s3, from_info, to_info, extra_args):
         if etag != cached_etag:
             raise ETagMismatchError(etag, cached_etag)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         return self.get_etag(self.s3, path_info.bucket, path_info.path)
 
     def _upload(self, from_file, to_info, name=None, no_progress_bar=False):
diff --git a/dvc/tree/ssh/__init__.py b/dvc/tree/ssh/__init__.py
index 26bee19388..20560f8e6f 100644
--- a/dvc/tree/ssh/__init__.py
+++ b/dvc/tree/ssh/__init__.py
@@ -233,7 +233,8 @@ def reflink(self, from_info, to_info):
         with self.ssh(from_info) as ssh:
             ssh.reflink(from_info.path, to_info.path)
 
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         if path_info.scheme != self.scheme:
             raise NotImplementedError
 
diff --git a/dvc/tree/webdav.py b/dvc/tree/webdav.py
index 115de31767..282851707e 100644
--- a/dvc/tree/webdav.py
+++ b/dvc/tree/webdav.py
@@ -131,7 +131,8 @@ def exists(self, path_info, use_dvcignore=True):
         return self._client.check(path_info.path)
 
     # Gets file hash 'etag'
-    def get_file_hash(self, path_info):
+    def get_file_hash(self, path_info, cmd=None):
+        assert not cmd, NotImplementedError
         # Use webdav client info method to get etag
         etag = self._client.info(path_info.path)["etag"].strip('"')
 
diff --git a/dvc/utils/__init__.py b/dvc/utils/__init__.py
index 8fece228ff..15966ba0ad 100644
--- a/dvc/utils/__init__.py
+++ b/dvc/utils/__init__.py
@@ -7,7 +7,9 @@
 import math
 import os
 import re
+import subprocess
 import sys
+import tempfile
 import time
 
 import colorama
@@ -43,8 +45,10 @@ def _fobj_md5(fobj, hash_md5, binary, progress_func=None):
             progress_func(len(data))
 
 
-def file_md5(fname, tree=None):
-    """ get the (md5 hexdigest, md5 digest) of a file """
+def file_md5(fname, tree=None, cmd=None):
+    """
+    Returns (md5_hexdigest, md5_digest) of `cmd file` (default: `cmd=cat`)
+    """
     from dvc.progress import Tqdm
     from dvc.istextfile import istextfile
 
@@ -58,6 +62,21 @@ def file_md5(fname, tree=None):
         open_func = open
 
     if exists_func(fname):
+        filtered = None
+        if cmd:
+            p = subprocess.Popen(
+                cmd.split() + [fname],
+                stdout=subprocess.PIPE,
+                stderr=subprocess.PIPE,
+            )
+            out, err = p.communicate()
+            if p.returncode != 0:
+                logger.error("filtering:%s %s", cmd, fname)
+                raise RuntimeError(err)
+            with tempfile.NamedTemporaryFile(delete=False) as fobj:
+                logger.debug("filtering:%s %s > %s", cmd, fname, fobj.name)
+                fobj.write(out)
+                fname = filtered = fobj.name
         hash_md5 = hashlib.md5()
         binary = not istextfile(fname, tree=tree)
         size = stat_func(fname).st_size
@@ -80,6 +99,10 @@ def file_md5(fname, tree=None):
             with open_func(fname, "rb") as fobj:
                 _fobj_md5(fobj, hash_md5, binary, pbar.update)
 
+        if filtered is not None:
+            from dvc.utils.fs import remove
+
+            remove(filtered)
         return (hash_md5.hexdigest(), hash_md5.digest())
 
     return (None, None)